Skip to content

GitLab

Unverified Commit d7067e44 authored by Wenwei Zhang's avatar Wenwei Zhang Committed by GitHub
Browse files

Bump version to v1.1.0rc2 

Bump to v1.1.0rc2
parents 28fe73d2 fb0e57e5
Show whitespace changes
Inline Side-by-side
Showing with 971 additions and 769 deletions
mmdet3d/datasets/det3d_dataset.py
View file @ d7067e44
# Copyright (c) OpenMMLab. All rights reserved.
import copy
import os
from os import path as osp
from typing import Callable, List, Optional, Union
from typing import Callable, List, Optional, Set, Union
import mmengine
import numpy as np
import torch
from mmengine.dataset import BaseDataset
from mmengine.logging import print_log
from terminaltables import AsciiTable
from mmdet3d.datasets import DATASETS
from mmdet3d.structures import get_box_type
......@@ -25,22 +29,22 @@ class Det3DDataset(BaseDataset):
ann_file (str): Annotation file path. Defaults to ''.
metainfo (dict, optional): Meta information for dataset, such as class
information. Defaults to None.
data_prefix (dict, optional): Prefix for training data. Defaults to
data_prefix (dict): Prefix for training data. Defaults to
dict(pts='velodyne', img='').
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input, it usually has following keys:
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used as input,
it usually has following keys:
- use_camera: bool
- use_lidar: bool
Defaults to `dict(use_lidar=True, use_camera=False)`
Defaults to dict(use_lidar=True, use_camera=False).
default_cam_key (str, optional): The default camera name adopted.
Defaults to None.
box_type_3d (str, optional): Type of 3D box of this dataset.
box_type_3d (str): Type of 3D box of this dataset.
Based on the `box_type_3d`, the dataset will encapsulate the box
to its original format then converted them to `box_type_3d`.
Defaults to 'LiDAR'. Available options includes:
Defaults to 'LiDAR' in this dataset. Available options includes:
- 'LiDAR': Box in LiDAR coordinates, usually for
outdoor point cloud 3d detection.
......@@ -48,16 +52,20 @@ class Det3DDataset(BaseDataset):
indoor point cloud 3d detection.
- 'Camera': Box in camera coordinates, usually
for vision-based 3d detection.
filter_empty_gt (bool, optional): Whether to filter the data with
empty GT. Defaults to True.
test_mode (bool, optional): Whether the dataset is in test mode.
filter_empty_gt (bool): Whether to filter the data with empty GT.
If it's set to be True, the example with empty annotations after
data pipeline will be dropped and a random example will be chosen
in `__getitem__`. Defaults to True.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
load_eval_anns (bool, optional): Whether to load annotations
in test_mode, the annotation will be save in `eval_ann_infos`,
which can be used in Evaluator. Defaults to True.
file_client_args (dict, optional): Configuration of file client.
load_eval_anns (bool): Whether to load annotations in test_mode,
the annotation will be save in `eval_ann_infos`, which can be
used in Evaluator. Defaults to True.
file_client_args (dict): Configuration of file client.
Defaults to dict(backend='disk').
show_ins_var (bool): For debug purpose. Whether to show variation
of the number of instances before and after through pipeline.
Defaults to False.
"""
def __init__(self,
......@@ -71,8 +79,9 @@ class Det3DDataset(BaseDataset):
box_type_3d: dict = 'LiDAR',
filter_empty_gt: bool = True,
test_mode: bool = False,
load_eval_anns=True,
load_eval_anns: bool = True,
file_client_args: dict = dict(backend='disk'),
show_ins_var: bool = False,
**kwargs) -> None:
# init file client
self.file_client = mmengine.FileClient(**file_client_args)
......@@ -94,24 +103,31 @@ class Det3DDataset(BaseDataset):
self.box_type_3d, self.box_mode_3d = get_box_type(box_type_3d)
if metainfo is not None and 'CLASSES' in metainfo:
# we allow to train on subset of self.METAINFO['CLASSES']
if metainfo is not None and 'classes' in metainfo:
# we allow to train on subset of self.METAINFO['classes']
# map unselected labels to -1
self.label_mapping = {
i: -1
for i in range(len(self.METAINFO['CLASSES']))
for i in range(len(self.METAINFO['classes']))
}
self.label_mapping[-1] = -1
for label_idx, name in enumerate(metainfo['CLASSES']):
ori_label = self.METAINFO['CLASSES'].index(name)
for label_idx, name in enumerate(metainfo['classes']):
ori_label = self.METAINFO['classes'].index(name)
self.label_mapping[ori_label] = label_idx
self.num_ins_per_cat = {name: 0 for name in metainfo['classes']}
else:
self.label_mapping = {
i: i
for i in range(len(self.METAINFO['CLASSES']))
for i in range(len(self.METAINFO['classes']))
}
self.label_mapping[-1] = -1
self.num_ins_per_cat = {
name: 0
for name in self.METAINFO['classes']
}
super().__init__(
ann_file=ann_file,
metainfo=metainfo,
......@@ -125,10 +141,25 @@ class Det3DDataset(BaseDataset):
self.metainfo['box_type_3d'] = box_type_3d
self.metainfo['label_mapping'] = self.label_mapping
# used for showing variation of the number of instances before and
# after through the pipeline
self.show_ins_var = show_ins_var
# show statistics of this dataset
print_log('-' * 30, 'current')
print_log(f'The length of the dataset: {len(self)}', 'current')
content_show = [['category', 'number']]
for cat_name, num in self.num_ins_per_cat.items():
content_show.append([cat_name, num])
table = AsciiTable(content_show)
print_log(
f'The number of instances per category in the dataset:\n{table.table}', # noqa: E501
'current')
def _remove_dontcare(self, ann_info: dict) -> dict:
"""Remove annotations that do not need to be cared.
-1 indicate dontcare in MMDet3d.
-1 indicates dontcare in MMDet3d.
Args:
ann_info (dict): Dict of annotation infos. The
......@@ -156,7 +187,7 @@ class Det3DDataset(BaseDataset):
index (int): Index of the annotation data to get.
Returns:
dict: annotation information.
dict: Annotation information.
"""
data_info = self.get_data_info(index)
# test model
......@@ -167,8 +198,8 @@ class Det3DDataset(BaseDataset):
return ann_info
def parse_ann_info(self, info: dict) -> Optional[dict]:
"""Process the `instances` in data info to `ann_info`
def parse_ann_info(self, info: dict) -> Union[dict, None]:
"""Process the `instances` in data info to `ann_info`.
In `Custom3DDataset`, we simply concatenate all the field
in `instances` to `np.ndarray`, you can do the specific
......@@ -179,7 +210,7 @@ class Det3DDataset(BaseDataset):
info (dict): Info dict.
Returns:
dict | None: Processed `ann_info`
dict or None: Processed `ann_info`.
"""
# add s or gt prefix for most keys after concat
# we only process 3d annotations here, the corresponding
......@@ -223,14 +254,20 @@ class Det3DDataset(BaseDataset):
ann_info[mapped_ann_name] = temp_anns
ann_info['instances'] = info['instances']
for label in ann_info['gt_labels_3d']:
if label != -1:
cat_name = self.metainfo['classes'][label]
self.num_ins_per_cat[cat_name] += 1
return ann_info
def parse_data_info(self, info: dict) -> dict:
"""Process the raw data info.
Convert all relative path of needed modality data file to
the absolute path. And process
the `instances` field to `ann_info` in training stage.
the absolute path. And process the `instances` field to
`ann_info` in training stage.
Args:
info (dict): Raw info dict.
......@@ -251,7 +288,7 @@ class Det3DDataset(BaseDataset):
if 'lidar_sweeps' in info:
for sweep in info['lidar_sweeps']:
file_suffix = sweep['lidar_points']['lidar_path'].split(
'/')[-1]
os.sep)[-1]
if 'samples' in sweep['lidar_points']['lidar_path']:
sweep['lidar_points']['lidar_path'] = osp.join(
self.data_prefix['pts'], file_suffix)
......@@ -291,7 +328,37 @@ class Det3DDataset(BaseDataset):
return info
def prepare_data(self, index: int) -> Optional[dict]:
def _show_ins_var(self, old_labels: np.ndarray,
new_labels: torch.Tensor) -> None:
"""Show variation of the number of instances before and after through
the pipeline.
Args:
old_labels (np.ndarray): The labels before through the pipeline.
new_labels (torch.Tensor): The labels after through the pipeline.
"""
ori_num_per_cat = dict()
for label in old_labels:
if label != -1:
cat_name = self.metainfo['classes'][label]
ori_num_per_cat[cat_name] = ori_num_per_cat.get(cat_name,
0) + 1
new_num_per_cat = dict()
for label in new_labels:
if label != -1:
cat_name = self.metainfo['classes'][label]
new_num_per_cat[cat_name] = new_num_per_cat.get(cat_name,
0) + 1
content_show = [['category', 'new number', 'ori number']]
for cat_name, num in ori_num_per_cat.items():
new_num = new_num_per_cat.get(cat_name, 0)
content_show.append([cat_name, new_num, num])
table = AsciiTable(content_show)
print_log(
'The number of instances per category after and before '
f'through pipeline:\n{table.table}', 'current')
def prepare_data(self, index: int) -> Union[dict, None]:
"""Data preparation for both training and testing stage.
Called by `__getitem__` of dataset.
......@@ -300,12 +367,12 @@ class Det3DDataset(BaseDataset):
index (int): Index for accessing the target data.
Returns:
dict | None: Data dict of the corresponding index.
dict or None: Data dict of the corresponding index.
"""
input_dict = self.get_data_info(index)
ori_input_dict = self.get_data_info(index)
# deepcopy here to avoid inplace modification in pipeline.
input_dict = copy.deepcopy(input_dict)
input_dict = copy.deepcopy(ori_input_dict)
# box_type_3d (str): 3D box type.
input_dict['box_type_3d'] = self.box_type_3d
......@@ -318,15 +385,29 @@ class Det3DDataset(BaseDataset):
return None
example = self.pipeline(input_dict)
if not self.test_mode and self.filter_empty_gt:
# after pipeline drop the example with empty annotations
# return None to random another in `__getitem__`
if example is None or len(
example['data_samples'].gt_instances_3d.labels_3d) == 0:
return None
if self.show_ins_var:
if 'ann_info' in ori_input_dict:
self._show_ins_var(
ori_input_dict['ann_info']['gt_labels_3d'],
example['data_samples'].gt_instances_3d.labels_3d)
else:
print_log(
"'ann_info' is not in the input dict. It's probably that "
'the data is not in training mode',
'current',
level=30)
return example
def get_cat_ids(self, idx: int) -> List[int]:
def get_cat_ids(self, idx: int) -> Set[int]:
"""Get category ids by index. Dataset wrapped by ClassBalancedDataset
must implement this method.
......
mmdet3d/datasets/kitti2d_dataset.py
View file @ d7067e44
......@@ -36,7 +36,7 @@ class Kitti2DDataset(Det3DDataset):
Defaults to False.
"""
CLASSES = ('car', 'pedestrian', 'cyclist')
classes = ('car', 'pedestrian', 'cyclist')
"""
Annotation format:
[
......@@ -90,7 +90,7 @@ class Kitti2DDataset(Det3DDataset):
self.data_infos = mmengine.load(ann_file)
self.cat2label = {
cat_name: i
for i, cat_name in enumerate(self.CLASSES)
for i, cat_name in enumerate(self.classes)
}
return self.data_infos
......@@ -122,7 +122,7 @@ class Kitti2DDataset(Det3DDataset):
difficulty = annos['difficulty']
# remove classes that is not needed
selected = self.keep_arrays_by_name(gt_names, self.CLASSES)
selected = self.keep_arrays_by_name(gt_names, self.classes)
gt_bboxes = gt_bboxes[selected]
gt_names = gt_names[selected]
difficulty = difficulty[selected]
......@@ -215,7 +215,7 @@ class Kitti2DDataset(Det3DDataset):
"""
from mmdet3d.structures.ops.transforms import bbox2result_kitti2d
sample_idx = [info['image']['image_idx'] for info in self.data_infos]
result_files = bbox2result_kitti2d(outputs, self.CLASSES, sample_idx,
result_files = bbox2result_kitti2d(outputs, self.classes, sample_idx,
out)
return result_files
......@@ -237,5 +237,5 @@ class Kitti2DDataset(Det3DDataset):
]), 'KITTI data set only evaluate bbox'
gt_annos = [info['annos'] for info in self.data_infos]
ap_result_str, ap_dict = kitti_eval(
gt_annos, result_files, self.CLASSES, eval_types=['bbox'])
gt_annos, result_files, self.classes, eval_types=['bbox'])
return ap_result_str, ap_dict
mmdet3d/datasets/kitti_dataset.py
View file @ d7067e44
......@@ -18,13 +18,13 @@ class KittiDataset(Det3DDataset):
Args:
data_root (str): Path of dataset root.
ann_file (str): Path of annotation file.
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input. Defaults to `dict(use_lidar=True)`.
default_cam_key (str, optional): The default camera name adopted.
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used as input.
Defaults to dict(use_lidar=True).
default_cam_key (str): The default camera name adopted.
Defaults to 'CAM2'.
box_type_3d (str, optional): Type of 3D box of this dataset.
box_type_3d (str): Type of 3D box of this dataset.
Based on the `box_type_3d`, the dataset will encapsulate the box
to its original format then converted them to `box_type_3d`.
Defaults to 'LiDAR' in this dataset. Available options includes:
......@@ -32,17 +32,28 @@ class KittiDataset(Det3DDataset):
- 'LiDAR': Box in LiDAR coordinates.
- 'Depth': Box in depth coordinates, usually for indoor dataset.
- 'Camera': Box in camera coordinates.
filter_empty_gt (bool, optional): Whether to filter empty GT.
Defaults to True.
test_mode (bool, optional): Whether the dataset is in test mode.
load_type (str): Type of loading mode. Defaults to 'frame_based'.
- 'frame_based': Load all of the instances in the frame.
- 'mv_image_based': Load all of the instances in the frame and need
to convert to the FOV-based data type to support image-based
detector.
- 'fov_image_based': Only load the instances inside the default
cam, and need to convert to the FOV-based data type to support
image-based detector.
filter_empty_gt (bool): Whether to filter the data with empty GT.
If it's set to be True, the example with empty annotations after
data pipeline will be dropped and a random example will be chosen
in `__getitem__`. Defaults to True.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
pcd_limit_range (list[float], optional): The range of point cloud
used to filter invalid predicted boxes.
pcd_limit_range (List[float]): The range of point cloud used to filter
invalid predicted boxes.
Defaults to [0, -40, -3, 70.4, 40, 0.0].
"""
# TODO: use full classes of kitti
METAINFO = {
'CLASSES': ('Pedestrian', 'Cyclist', 'Car', 'Van', 'Truck',
'classes': ('Pedestrian', 'Cyclist', 'Car', 'Van', 'Truck',
'Person_sitting', 'Tram', 'Misc')
}
......@@ -52,7 +63,7 @@ class KittiDataset(Det3DDataset):
pipeline: List[Union[dict, Callable]] = [],
modality: dict = dict(use_lidar=True),
default_cam_key: str = 'CAM2',
task: str = 'lidar_det',
load_type: str = 'frame_based',
box_type_3d: str = 'LiDAR',
filter_empty_gt: bool = True,
test_mode: bool = False,
......@@ -60,8 +71,9 @@ class KittiDataset(Det3DDataset):
**kwargs) -> None:
self.pcd_limit_range = pcd_limit_range
assert task in ('lidar_det', 'mono_det')
self.task = task
assert load_type in ('frame_based', 'mv_image_based',
'fov_image_based')
self.load_type = load_type
super().__init__(
data_root=data_root,
ann_file=ann_file,
......@@ -111,7 +123,7 @@ class KittiDataset(Det3DDataset):
info['plane'] = plane_lidar
if self.task == 'mono_det':
if self.load_type == 'fov_image_based' and self.load_eval_anns:
info['instances'] = info['cam_instances'][self.default_cam_key]
info = super().parse_data_info(info)
......@@ -119,13 +131,13 @@ class KittiDataset(Det3DDataset):
return info
def parse_ann_info(self, info: dict) -> dict:
"""Get annotation info according to the given index.
"""Process the `instances` in data info to `ann_info`.
Args:
info (dict): Data information of single data sample.
Returns:
dict: annotation information consists of the following keys:
dict: Annotation information consists of the following keys:
- gt_bboxes_3d (:obj:`LiDARInstance3DBoxes`):
3D ground truth bboxes.
......@@ -142,7 +154,7 @@ class KittiDataset(Det3DDataset):
ann_info['gt_bboxes_3d'] = np.zeros((0, 7), dtype=np.float32)
ann_info['gt_labels_3d'] = np.zeros(0, dtype=np.int64)
if self.task == 'mono_det':
if self.load_type in ['fov_image_based', 'mv_image_based']:
ann_info['gt_bboxes'] = np.zeros((0, 4), dtype=np.float32)
ann_info['gt_bboxes_labels'] = np.array(0, dtype=np.int64)
ann_info['centers_2d'] = np.zeros((0, 2), dtype=np.float32)
......
mmdet3d/datasets/lyft_dataset.py
View file @ d7067e44
......@@ -21,10 +21,10 @@ class LyftDataset(Det3DDataset):
Args:
data_root (str): Path of dataset root.
ann_file (str): Path of annotation file.
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input. Defaults to dict(use_camera=False, use_lidar=True).
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used as input.
Defaults to dict(use_camera=False, use_lidar=True).
box_type_3d (str): Type of 3D box of this dataset.
Based on the `box_type_3d`, the dataset will encapsulate the box
to its original format then converted them to `box_type_3d`.
......@@ -33,14 +33,16 @@ class LyftDataset(Det3DDataset):
- 'LiDAR': Box in LiDAR coordinates.
- 'Depth': Box in depth coordinates, usually for indoor dataset.
- 'Camera': Box in camera coordinates.
filter_empty_gt (bool, optional): Whether to filter empty GT.
Defaults to True.
test_mode (bool, optional): Whether the dataset is in test mode.
filter_empty_gt (bool): Whether to filter the data with empty GT.
If it's set to be True, the example with empty annotations after
data pipeline will be dropped and a random example will be chosen
in `__getitem__`. Defaults to True.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
"""
METAINFO = {
'CLASSES':
'classes':
('car', 'truck', 'bus', 'emergency_vehicle', 'other_vehicle',
'motorcycle', 'bicycle', 'pedestrian', 'animal')
}
......@@ -66,13 +68,13 @@ class LyftDataset(Det3DDataset):
**kwargs)
def parse_ann_info(self, info: dict) -> dict:
"""Get annotation info according to the given index.
"""Process the `instances` in data info to `ann_info`.
Args:
info (dict): Data information of single data sample.
Returns:
dict: annotation information consists of the following keys:
dict: Annotation information consists of the following keys:
- gt_bboxes_3d (:obj:`LiDARInstance3DBoxes`):
3D ground truth bboxes.
......
mmdet3d/datasets/nuscenes_dataset.py
View file @ d7067e44
......@@ -22,9 +22,8 @@ class NuScenesDataset(Det3DDataset):
Args:
data_root (str): Path of dataset root.
ann_file (str): Path of annotation file.
task (str, optional): Detection task. Defaults to 'lidar_det'.
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
pipeline (list[dict]): Pipeline used for data processing.
Defaults to [].
box_type_3d (str): Type of 3D box of this dataset.
Based on the `box_type_3d`, the dataset will encapsulate the box
to its original format then converted them to `box_type_3d`.
......@@ -33,20 +32,31 @@ class NuScenesDataset(Det3DDataset):
- 'LiDAR': Box in LiDAR coordinates.
- 'Depth': Box in depth coordinates, usually for indoor dataset.
- 'Camera': Box in camera coordinates.
modality (dict, optional): Modality to specify the sensor data used
as input. Defaults to dict(use_camera=False, use_lidar=True).
filter_empty_gt (bool, optional): Whether to filter empty GT.
Defaults to True.
test_mode (bool, optional): Whether the dataset is in test mode.
load_type (str): Type of loading mode. Defaults to 'frame_based'.
- 'frame_based': Load all of the instances in the frame.
- 'mv_image_based': Load all of the instances in the frame and need
to convert to the FOV-based data type to support image-based
detector.
- 'fov_image_based': Only load the instances inside the default
cam, and need to convert to the FOV-based data type to support
image-based detector.
modality (dict): Modality to specify the sensor data used as input.
Defaults to dict(use_camera=False, use_lidar=True).
filter_empty_gt (bool): Whether to filter the data with empty GT.
If it's set to be True, the example with empty annotations after
data pipeline will be dropped and a random example will be chosen
in `__getitem__`. Defaults to True.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
with_velocity (bool, optional): Whether to include velocity prediction
with_velocity (bool): Whether to include velocity prediction
into the experiments. Defaults to True.
use_valid_flag (bool, optional): Whether to use `use_valid_flag` key
use_valid_flag (bool): Whether to use `use_valid_flag` key
in the info file as mask to filter gt_boxes and gt_names.
Defaults to False.
"""
METAINFO = {
'CLASSES':
'classes':
('car', 'truck', 'trailer', 'bus', 'construction_vehicle', 'bicycle',
'motorcycle', 'pedestrian', 'traffic_cone', 'barrier'),
'version':
......@@ -56,9 +66,9 @@ class NuScenesDataset(Det3DDataset):
def __init__(self,
data_root: str,
ann_file: str,
task: str = 'lidar_det',
pipeline: List[Union[dict, Callable]] = [],
box_type_3d: str = 'LiDAR',
load_type: str = 'frame_based',
modality: dict = dict(
use_camera=False,
use_lidar=True,
......@@ -72,8 +82,9 @@ class NuScenesDataset(Det3DDataset):
self.with_velocity = with_velocity
# TODO: Redesign multi-view data process in the future
assert task in ('lidar_det', 'mono_det', 'multi-view_det')
self.task = task
assert load_type in ('frame_based', 'mv_image_based',
'fov_image_based')
self.load_type = load_type
assert box_type_3d.lower() in ('lidar', 'camera')
super().__init__(
......@@ -108,13 +119,13 @@ class NuScenesDataset(Det3DDataset):
return filtered_annotations
def parse_ann_info(self, info: dict) -> dict:
"""Get annotation info according to the given index.
"""Process the `instances` in data info to `ann_info`.
Args:
info (dict): Data information of single data sample.
Returns:
dict: annotation information consists of the following keys:
dict: Annotation information consists of the following keys:
- gt_bboxes_3d (:obj:`LiDARInstance3DBoxes`):
3D ground truth bboxes.
......@@ -142,7 +153,7 @@ class NuScenesDataset(Det3DDataset):
ann_info['gt_bboxes_3d'] = np.zeros((0, 7), dtype=np.float32)
ann_info['gt_labels_3d'] = np.zeros(0, dtype=np.int64)
if self.task == 'mono3d':
if self.load_type in ['fov_image_based', 'mv_image_based']:
ann_info['gt_bboxes'] = np.zeros((0, 4), dtype=np.float32)
ann_info['gt_bboxes_labels'] = np.array(0, dtype=np.int64)
ann_info['attr_labels'] = np.array(0, dtype=np.int64)
......@@ -152,7 +163,7 @@ class NuScenesDataset(Det3DDataset):
# the nuscenes box center is [0.5, 0.5, 0.5], we change it to be
# the same as KITTI (0.5, 0.5, 0)
# TODO: Unify the coordinates
if self.task == 'mono_det':
if self.load_type in ['fov_image_based', 'mv_image_based']:
gt_bboxes_3d = CameraInstance3DBoxes(
ann_info['gt_bboxes_3d'],
box_dim=ann_info['gt_bboxes_3d'].shape[-1],
......@@ -167,7 +178,7 @@ class NuScenesDataset(Det3DDataset):
return ann_info
def parse_data_info(self, info: dict) -> dict:
def parse_data_info(self, info: dict) -> Union[List[dict], dict]:
"""Process the raw data info.
The only difference with it in `Det3DDataset`
......@@ -177,10 +188,10 @@ class NuScenesDataset(Det3DDataset):
info (dict): Raw info dict.
Returns:
dict: Has `ann_info` in training stage. And
List[dict] or dict: Has `ann_info` in training stage. And
all path has been converted to absolute path.
"""
if self.task == 'mono_det':
if self.load_type == 'mv_image_based':
data_list = []
if self.modality['use_lidar']:
info['lidar_points']['lidar_path'] = \
......
mmdet3d/datasets/s3dis_dataset.py
View file @ d7067e44
# Copyright (c) OpenMMLab. All rights reserved.
from os import path as osp
from typing import Callable, List, Optional, Union
from typing import Any, Callable, List, Optional, Tuple, Union
import numpy as np
......@@ -8,7 +8,6 @@ from mmdet3d.registry import DATASETS
from mmdet3d.structures import DepthInstance3DBoxes
from .det3d_dataset import Det3DDataset
from .seg3d_dataset import Seg3DDataset
from .transforms import Compose
@DATASETS.register_module()
......@@ -19,138 +18,132 @@ class S3DISDataset(Det3DDataset):
often train on 5 of them and test on the remaining one. The one for
test is Area_5 as suggested in `GSDN <https://arxiv.org/abs/2006.12356>`_.
To concatenate 5 areas during training
`mmdet.datasets.dataset_wrappers.ConcatDataset` should be used.
`mmengine.datasets.dataset_wrappers.ConcatDataset` should be used.
Args:
data_root (str): Path of dataset root.
ann_file (str): Path of annotation file.
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
classes (tuple[str], optional): Classes used in the dataset.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input. Defaults to None.
box_type_3d (str, optional): Type of 3D box of this dataset.
metainfo (dict, optional): Meta information for dataset, such as class
information. Defaults to None.
data_prefix (dict): Prefix for data. Defaults to
dict(pts='points',
pts_instance_mask='instance_mask',
pts_semantic_mask='semantic_mask').
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used as input.
Defaults to dict(use_camera=False, use_lidar=True).
box_type_3d (str): Type of 3D box of this dataset.
Based on the `box_type_3d`, the dataset will encapsulate the box
to its original format then converted them to `box_type_3d`.
Defaults to 'Depth' in this dataset. Available options includes
Defaults to 'Depth' in this dataset. Available options includes:
- 'LiDAR': Box in LiDAR coordinates.
- 'Depth': Box in depth coordinates, usually for indoor dataset.
- 'Camera': Box in camera coordinates.
filter_empty_gt (bool, optional): Whether to filter empty GT.
Defaults to True.
test_mode (bool, optional): Whether the dataset is in test mode.
filter_empty_gt (bool): Whether to filter the data with empty GT.
If it's set to be True, the example with empty annotations after
data pipeline will be dropped and a random example will be chosen
in `__getitem__`. Defaults to True.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
"""
CLASSES = ('table', 'chair', 'sofa', 'bookcase', 'board')
METAINFO = {
'classes': ('table', 'chair', 'sofa', 'bookcase', 'board'),
# the valid ids of segmentation annotations
'seg_valid_class_ids': (7, 8, 9, 10, 11),
'seg_all_class_ids': tuple(range(1, 14)) # possibly with 'stair' class
}
def __init__(self,
data_root,
ann_file,
pipeline=None,
classes=None,
modality=None,
box_type_3d='Depth',
filter_empty_gt=True,
test_mode=False,
*kwargs):
data_root: str,
ann_file: str,
metainfo: Optional[dict] = None,
data_prefix: dict = dict(
pts='points',
pts_instance_mask='instance_mask',
pts_semantic_mask='semantic_mask'),
pipeline: List[Union[dict, Callable]] = [],
modality: dict = dict(use_camera=False, use_lidar=True),
box_type_3d: str = 'Depth',
filter_empty_gt: bool = True,
test_mode: bool = False,
**kwargs) -> None:
# construct seg_label_mapping for semantic mask
seg_max_cat_id = len(self.METAINFO['seg_all_class_ids'])
seg_valid_cat_ids = self.METAINFO['seg_valid_class_ids']
neg_label = len(seg_valid_cat_ids)
seg_label_mapping = np.ones(
seg_max_cat_id + 1, dtype=np.int) * neg_label
for cls_idx, cat_id in enumerate(seg_valid_cat_ids):
seg_label_mapping[cat_id] = cls_idx
self.seg_label_mapping = seg_label_mapping
super().__init__(
data_root=data_root,
ann_file=ann_file,
metainfo=metainfo,
data_prefix=data_prefix,
pipeline=pipeline,
classes=classes,
modality=modality,
box_type_3d=box_type_3d,
filter_empty_gt=filter_empty_gt,
test_mode=test_mode,
*kwargs)
**kwargs)
self.metainfo['seg_label_mapping'] = self.seg_label_mapping
assert 'use_camera' in self.modality and \
'use_lidar' in self.modality
assert self.modality['use_camera'] or self.modality['use_lidar']
def get_ann_info(self, index):
"""Get annotation info according to the given index.
def parse_data_info(self, info: dict) -> dict:
"""Process the raw data info.
Args:
index (int): Index of the annotation data to get.
info (dict): Raw info dict.
Returns:
dict: annotation information consists of the following keys:
- gt_bboxes_3d (:obj:`DepthInstance3DBoxes`):
3D ground truth bboxes
- gt_labels_3d (np.ndarray): Labels of ground truths.
- pts_instance_mask_path (str): Path of instance masks.
- pts_semantic_mask_path (str): Path of semantic masks.
dict: Has `ann_info` in training stage. And
all path has been converted to absolute path.
"""
# Use index to get the annos, thus the evalhook could also use this api
info = self.data_infos[index]
if info['annos']['gt_num'] != 0:
gt_bboxes_3d = info['annos']['gt_boxes_upright_depth'].astype(
np.float32) # k, 6
gt_labels_3d = info['annos']['class'].astype(np.int64)
else:
gt_bboxes_3d = np.zeros((0, 6), dtype=np.float32)
gt_labels_3d = np.zeros((0, ), dtype=np.int64)
# to target box structure
gt_bboxes_3d = DepthInstance3DBoxes(
gt_bboxes_3d,
box_dim=gt_bboxes_3d.shape[-1],
with_yaw=False,
origin=(0.5, 0.5, 0.5)).convert_to(self.box_mode_3d)
pts_instance_mask_path = osp.join(self.data_root,
info['pts_instance_mask_path'] = osp.join(
self.data_prefix.get('pts_instance_mask', ''),
info['pts_instance_mask_path'])
pts_semantic_mask_path = osp.join(self.data_root,
info['pts_semantic_mask_path'] = osp.join(
self.data_prefix.get('pts_semantic_mask', ''),
info['pts_semantic_mask_path'])
anns_results = dict(
gt_bboxes_3d=gt_bboxes_3d,
gt_labels_3d=gt_labels_3d,
pts_instance_mask_path=pts_instance_mask_path,
pts_semantic_mask_path=pts_semantic_mask_path)
return anns_results
info = super().parse_data_info(info)
# only be used in `PointSegClassMapping` in pipeline
# to map original semantic class to valid category ids.
info['seg_label_mapping'] = self.seg_label_mapping
return info
def get_data_info(self, index):
"""Get data info according to the given index.
def parse_ann_info(self, info: dict) -> dict:
"""Process the `instances` in data info to `ann_info`.
Args:
index (int): Index of the sample data to get.
info (dict): Info dict.
Returns:
dict: Data information that will be passed to the data
preprocessing transforms. It includes the following keys:
- pts_filename (str): Filename of point clouds.
- file_name (str): Filename of point clouds.
- ann_info (dict): Annotation info.
dict: Processed `ann_info`.
"""
info = self.data_infos[index]
pts_filename = osp.join(self.data_root, info['pts_path'])
input_dict = dict(pts_filename=pts_filename)
ann_info = super().parse_ann_info(info)
# empty gt
if ann_info is None:
ann_info = dict()
ann_info['gt_bboxes_3d'] = np.zeros((0, 6), dtype=np.float32)
ann_info['gt_labels_3d'] = np.zeros((0, ), dtype=np.int64)
# to target box structure
if not self.test_mode:
annos = self.get_ann_info(index)
input_dict['ann_info'] = annos
if self.filter_empty_gt and ~(annos['gt_labels_3d'] != -1).any():
return None
return input_dict
def _build_default_pipeline(self):
"""Build the default pipeline for this dataset."""
pipeline = [
dict(
type='LoadPointsFromFile',
coord_type='DEPTH',
shift_height=False,
load_dim=6,
use_dim=[0, 1, 2, 3, 4, 5]),
dict(
type='DefaultFormatBundle3D',
class_names=self.CLASSES,
with_label=False),
dict(type='Collect3D', keys=['points'])
]
return Compose(pipeline)
ann_info['gt_bboxes_3d'] = DepthInstance3DBoxes(
ann_info['gt_bboxes_3d'],
box_dim=ann_info['gt_bboxes_3d'].shape[-1],
with_yaw=False,
origin=(0.5, 0.5, 0.5)).convert_to(self.box_mode_3d)
return ann_info
class _S3DISSegDataset(Seg3DDataset):
......@@ -166,30 +159,31 @@ class _S3DISSegDataset(Seg3DDataset):
wrapper to concat all the provided data in different areas.
Args:
data_root (str): Path of dataset root.
ann_file (str): Path of annotation file.
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
classes (tuple[str], optional): Classes used in the dataset.
Defaults to None.
palette (list[list[int]], optional): The palette of segmentation map.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input. Defaults to None.
test_mode (bool, optional): Whether the dataset is in test mode.
Defaults to False.
data_root (str, optional): Path of dataset root, Defaults to None.
ann_file (str): Path of annotation file. Defaults to ''.
metainfo (dict, optional): Meta information for dataset, such as class
information. Defaults to None.
data_prefix (dict): Prefix for training data. Defaults to
dict(pts='points', pts_instance_mask='', pts_semantic_mask='').
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used as input.
Defaults to dict(use_lidar=True, use_camera=False).
ignore_index (int, optional): The label index to be ignored, e.g.
unannotated points. If None is given, set to len(self.CLASSES).
unannotated points. If None is given, set to len(self.classes) to
be consistent with PointSegClassMapping function in pipeline.
Defaults to None.
scene_idxs (np.ndarray | str, optional): Precomputed index to load
scene_idxs (np.ndarray or str, optional): Precomputed index to load
data. For scenes with many points, we may sample it several times.
Defaults to None.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
"""
METAINFO = {
'CLASSES':
'classes':
('ceiling', 'floor', 'wall', 'beam', 'column', 'window', 'door',
'table', 'chair', 'sofa', 'bookcase', 'board', 'clutter'),
'PALETTE': [[0, 255, 0], [0, 0, 255], [0, 255, 255], [255, 255, 0],
'palette': [[0, 255, 0], [0, 0, 255], [0, 255, 255], [255, 255, 0],
[255, 0, 255], [100, 100, 255], [200, 200, 100],
[170, 120, 200], [255, 0, 0], [200, 100, 100],
[10, 200, 100], [200, 200, 200], [50, 50, 50]],
......@@ -204,12 +198,12 @@ class _S3DISSegDataset(Seg3DDataset):
ann_file: str = '',
metainfo: Optional[dict] = None,
data_prefix: dict = dict(
pts='points', img='', instance_mask='', semantic_mask=''),
pts='points', pts_instance_mask='', pts_semantic_mask=''),
pipeline: List[Union[dict, Callable]] = [],
modality: dict = dict(use_lidar=True, use_camera=False),
ignore_index=None,
scene_idxs=None,
test_mode=False,
ignore_index: Optional[int] = None,
scene_idxs: Optional[Union[np.ndarray, str]] = None,
test_mode: bool = False,
**kwargs) -> None:
super().__init__(
data_root=data_root,
......@@ -223,7 +217,8 @@ class _S3DISSegDataset(Seg3DDataset):
test_mode=test_mode,
**kwargs)
def get_scene_idxs(self, scene_idxs):
def get_scene_idxs(self, scene_idxs: Union[np.ndarray, str,
None]) -> np.ndarray:
"""Compute scene_idxs for data sampling.
We sample more times for scenes with more points.
......@@ -250,37 +245,40 @@ class S3DISSegDataset(_S3DISSegDataset):
data downloading.
Args:
data_root (str): Path of dataset root.
ann_files (list[str]): Path of several annotation files.
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
classes (tuple[str], optional): Classes used in the dataset.
Defaults to None.
palette (list[list[int]], optional): The palette of segmentation map.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input. Defaults to None.
test_mode (bool, optional): Whether the dataset is in test mode.
Defaults to False.
data_root (str, optional): Path of dataset root. Defaults to None.
ann_files (List[str]): Path of several annotation files.
Defaults to ''.
metainfo (dict, optional): Meta information for dataset, such as class
information. Defaults to None.
data_prefix (dict): Prefix for training data. Defaults to
dict(pts='points', pts_instance_mask='', pts_semantic_mask='').
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used as input.
Defaults to dict(use_lidar=True, use_camera=False).
ignore_index (int, optional): The label index to be ignored, e.g.
unannotated points. If None is given, set to len(self.CLASSES).
unannotated points. If None is given, set to len(self.classes) to
be consistent with PointSegClassMapping function in pipeline.
Defaults to None.
scene_idxs (list[np.ndarray] | list[str], optional): Precomputed index
to load data. For scenes with many points, we may sample it several
times. Defaults to None.
scene_idxs (List[np.ndarray] | List[str], optional): Precomputed index
to load data. For scenes with many points, we may sample it
several times. Defaults to None.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
"""
def __init__(self,
data_root: Optional[str] = None,
ann_files: str = '',
ann_files: List[str] = '',
metainfo: Optional[dict] = None,
data_prefix: dict = dict(
pts='points', img='', instance_mask='', semantic_mask=''),
pts='points', pts_instance_mask='', pts_semantic_mask=''),
pipeline: List[Union[dict, Callable]] = [],
modality: dict = dict(use_lidar=True, use_camera=False),
ignore_index=None,
scene_idxs=None,
test_mode=False,
ignore_index: Optional[int] = None,
scene_idxs: Optional[Union[List[np.ndarray],
List[str]]] = None,
test_mode: bool = False,
**kwargs) -> None:
# make sure that ann_files and scene_idxs have same length
......@@ -298,7 +296,6 @@ class S3DISSegDataset(_S3DISSegDataset):
ignore_index=ignore_index,
scene_idxs=scene_idxs[0],
test_mode=test_mode,
serialize_data=False,
**kwargs)
datasets = [
......@@ -312,56 +309,44 @@ class S3DISSegDataset(_S3DISSegDataset):
ignore_index=ignore_index,
scene_idxs=scene_idxs[i],
test_mode=test_mode,
serialize_data=False,
**kwargs) for i in range(len(ann_files))
]
# data_list and scene_idxs need to be concat
self.concat_data_list([dst.data_list for dst in datasets])
self.concat_scene_idxs([dst.scene_idxs for dst in datasets])
# set group flag for the sampler
if not self.test_mode:
self._set_group_flag()
def concat_data_list(self, data_lists):
def concat_data_list(self, data_lists: List[List[dict]]) -> None:
"""Concat data_list from several datasets to form self.data_list.
Args:
data_lists (list[list[dict]])
data_lists (List[List[dict]]): List of dict containing
annotation information.
"""
self.data_list = [
data for data_list in data_lists for data in data_list
]
def concat_scene_idxs(self, scene_idxs):
"""Concat scene_idxs from several datasets to form self.scene_idxs.
Needs to manually add offset to scene_idxs[1, 2, ...].
Args:
scene_idxs (list[np.ndarray])
"""
self.scene_idxs = np.array([], dtype=np.int32)
offset = 0
for one_scene_idxs in scene_idxs:
self.scene_idxs = np.concatenate(
[self.scene_idxs, one_scene_idxs + offset]).astype(np.int32)
offset = np.unique(self.scene_idxs).max() + 1
@staticmethod
def _duplicate_to_list(x, num):
def _duplicate_to_list(x: Any, num: int) -> list:
"""Repeat x `num` times to form a list."""
return [x for _ in range(num)]
def _check_ann_files(self, ann_file):
def _check_ann_files(
self, ann_file: Union[List[str], Tuple[str], str]) -> List[str]:
"""Make ann_files as list/tuple."""
# ann_file could be str
if not isinstance(ann_file, (list, tuple)):
ann_file = self._duplicate_to_list(ann_file, 1)
return ann_file
def _check_scene_idxs(self, scene_idx, num):
def _check_scene_idxs(self, scene_idx: Union[str, List[Union[list, tuple,
np.ndarray]],
List[str], None],
num: int) -> List[np.ndarray]:
"""Make scene_idxs as list/tuple."""
if scene_idx is None:
return self._duplicate_to_list(scene_idx, num)
......
mmdet3d/datasets/scannet_dataset.py
View file @ d7067e44
......@@ -26,13 +26,13 @@ class ScanNetDataset(Det3DDataset):
metainfo (dict, optional): Meta information for dataset, such as class
information. Defaults to None.
data_prefix (dict): Prefix for data. Defaults to
`dict(pts='points',
pts_isntance_mask='instance_mask',
pts_semantic_mask='semantic_mask')`.
pipeline (list[dict]): Pipeline used for data processing.
Defaults to None.
modality (dict): Modality to specify the sensor data used
as input. Defaults to None.
dict(pts='points',
pts_instance_mask='instance_mask',
pts_semantic_mask='semantic_mask').
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used as input.
Defaults to dict(use_camera=False, use_lidar=True).
box_type_3d (str): Type of 3D box of this dataset.
Based on the `box_type_3d`, the dataset will encapsulate the box
to its original format then converted them to `box_type_3d`.
......@@ -41,13 +41,15 @@ class ScanNetDataset(Det3DDataset):
- 'LiDAR': Box in LiDAR coordinates.
- 'Depth': Box in depth coordinates, usually for indoor dataset.
- 'Camera': Box in camera coordinates.
filter_empty_gt (bool): Whether to filter empty GT.
Defaults to True.
filter_empty_gt (bool): Whether to filter the data with empty GT.
If it's set to be True, the example with empty annotations after
data pipeline will be dropped and a random example will be chosen
in `__getitem__`. Defaults to True.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
"""
METAINFO = {
'CLASSES':
'classes':
('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', 'window',
'bookshelf', 'picture', 'counter', 'desk', 'curtain', 'refrigerator',
'showercurtrain', 'toilet', 'sink', 'bathtub', 'garbagebin'),
......@@ -71,7 +73,7 @@ class ScanNetDataset(Det3DDataset):
box_type_3d: str = 'Depth',
filter_empty_gt: bool = True,
test_mode: bool = False,
**kwargs):
**kwargs) -> None:
# construct seg_label_mapping for semantic mask
seg_max_cat_id = len(self.METAINFO['seg_all_class_ids'])
......@@ -128,8 +130,8 @@ class ScanNetDataset(Det3DDataset):
info (dict): Raw info dict.
Returns:
dict: Data information that will be passed to the data
preprocessing transforms. It includes the following keys:
dict: Has `ann_info` in training stage. And
all path has been converted to absolute path.
"""
info['axis_align_matrix'] = self._get_axis_align_matrix(info)
info['pts_instance_mask_path'] = osp.join(
......@@ -146,13 +148,13 @@ class ScanNetDataset(Det3DDataset):
return info
def parse_ann_info(self, info: dict) -> dict:
"""Process the `instances` in data info to `ann_info`
"""Process the `instances` in data info to `ann_info`.
Args:
info (dict): Info dict.
Returns:
dict: Processed `ann_info`
dict: Processed `ann_info`.
"""
ann_info = super().parse_ann_info(info)
# empty gt
......@@ -181,32 +183,36 @@ class ScanNetSegDataset(Seg3DDataset):
for data downloading.
Args:
data_root (str): Path of dataset root.
ann_file (str): Path of annotation file.
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
classes (tuple[str], optional): Classes used in the dataset.
Defaults to None.
palette (list[list[int]], optional): The palette of segmentation map.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input. Defaults to None.
test_mode (bool, optional): Whether the dataset is in test mode.
Defaults to False.
data_root (str, optional): Path of dataset root. Defaults to None.
ann_file (str): Path of annotation file. Defaults to ''.
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
metainfo (dict, optional): Meta information for dataset, such as class
information. Defaults to None.
data_prefix (dict): Prefix for training data. Defaults to
dict(pts='points',
img='',
pts_instance_mask='',
pts_semantic_mask='').
modality (dict): Modality to specify the sensor data used as input.
Defaults to dict(use_lidar=True, use_camera=False).
ignore_index (int, optional): The label index to be ignored, e.g.
unannotated points. If None is given, set to len(self.CLASSES).
unannotated points. If None is given, set to len(self.classes) to
be consistent with PointSegClassMapping function in pipeline.
Defaults to None.
scene_idxs (np.ndarray | str, optional): Precomputed index to load
scene_idxs (np.ndarray or str, optional): Precomputed index to load
data. For scenes with many points, we may sample it several times.
Defaults to None.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
"""
METAINFO = {
'CLASSES':
'classes':
('wall', 'floor', 'cabinet', 'bed', 'chair', 'sofa', 'table', 'door',
'window', 'bookshelf', 'picture', 'counter', 'desk', 'curtain',
'refrigerator', 'showercurtrain', 'toilet', 'sink', 'bathtub',
'otherfurniture'),
'PALETTE': [
'palette': [
[174, 199, 232],
[152, 223, 138],
[31, 119, 180],
......@@ -239,12 +245,15 @@ class ScanNetSegDataset(Seg3DDataset):
ann_file: str = '',
metainfo: Optional[dict] = None,
data_prefix: dict = dict(
pts='points', img='', instance_mask='', semantic_mask=''),
pts='points',
img='',
pts_instance_mask='',
pts_semantic_mask=''),
pipeline: List[Union[dict, Callable]] = [],
modality: dict = dict(use_lidar=True, use_camera=False),
ignore_index=None,
scene_idxs=None,
test_mode=False,
ignore_index: Optional[int] = None,
scene_idxs: Optional[Union[np.ndarray, str]] = None,
test_mode: bool = False,
**kwargs) -> None:
super().__init__(
data_root=data_root,
......@@ -258,7 +267,8 @@ class ScanNetSegDataset(Seg3DDataset):
test_mode=test_mode,
**kwargs)
def get_scene_idxs(self, scene_idxs):
def get_scene_idxs(self, scene_idxs: Union[np.ndarray, str,
None]) -> np.ndarray:
"""Compute scene_idxs for data sampling.
We sample more times for scenes with more points.
......@@ -275,11 +285,11 @@ class ScanNetSegDataset(Seg3DDataset):
class ScanNetInstanceSegDataset(Seg3DDataset):
METAINFO = {
'CLASSES':
'classes':
('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', 'window',
'bookshelf', 'picture', 'counter', 'desk', 'curtain', 'refrigerator',
'showercurtrain', 'toilet', 'sink', 'bathtub', 'garbagebin'),
'PLATTE': [
'palette': [
[174, 199, 232],
[152, 223, 138],
[31, 119, 180],
......@@ -312,13 +322,16 @@ class ScanNetInstanceSegDataset(Seg3DDataset):
ann_file: str = '',
metainfo: Optional[dict] = None,
data_prefix: dict = dict(
pts='points', img='', instance_mask='', semantic_mask=''),
pts='points',
img='',
pts_instance_mask='',
pts_semantic_mask=''),
pipeline: List[Union[dict, Callable]] = [],
modality: dict = dict(use_lidar=True, use_camera=False),
test_mode=False,
ignore_index=None,
scene_idxs=None,
file_client_args=dict(backend='disk'),
test_mode: bool = False,
ignore_index: Optional[int] = None,
scene_idxs: Optional[Union[np.ndarray, str]] = None,
file_client_args: dict = dict(backend='disk'),
**kwargs) -> None:
super().__init__(
data_root=data_root,
......
mmdet3d/datasets/seg3d_dataset.py
View file @ d7067e44
# Copyright (c) OpenMMLab. All rights reserved.
from os import path as osp
from typing import Callable, Dict, List, Optional, Sequence, Union
from typing import Callable, List, Optional, Sequence, Union
import mmengine
import numpy as np
......@@ -16,40 +16,45 @@ class Seg3DDataset(BaseDataset):
This is the base dataset of ScanNet, S3DIS and SemanticKITTI dataset.
Args:
data_root (str): Path of dataset root.
ann_file (str): Path of annotation file.
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
data_root (str, optional): Path of dataset root. Defaults to None.
ann_file (str): Path of annotation file. Defaults to ''.
metainfo (dict, optional): Meta information for dataset, such as class
information. Defaults to None.
data_prefix (dict, optional): Prefix for training data. Defaults to
dict(pts='velodyne', img='', instance_mask='', semantic_mask='').
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input, it usually has following keys.
data_prefix (dict): Prefix for training data. Defaults to
dict(pts='points',
img='',
pts_instance_mask='',
pts_semantic_mask='').
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used
as input, it usually has following keys:
- use_camera: bool
- use_lidar: bool
Defaults to `dict(use_lidar=True, use_camera=False)`
test_mode (bool, optional): Whether the dataset is in test mode.
Defaults to False.
Defaults to dict(use_lidar=True, use_camera=False).
ignore_index (int, optional): The label index to be ignored, e.g.
unannotated points. If None is given, set to len(self.CLASSES) to
unannotated points. If None is given, set to len(self.classes) to
be consistent with PointSegClassMapping function in pipeline.
Defaults to None.
scene_idxs (np.ndarray | str, optional): Precomputed index to load
scene_idxs (np.ndarray or str, optional): Precomputed index to load
data. For scenes with many points, we may sample it several times.
Defaults to None.
load_eval_anns (bool): Whether to load annotations
in test_mode, the annotation will be save in
`eval_ann_infos`, which can be use in Evaluator.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
serialize_data (bool): Whether to hold memory using serialized objects,
when enabled, data loader workers can use shared RAM from master
process instead of making a copy.
Defaults to False for 3D Segmentation datasets.
load_eval_anns (bool): Whether to load annotations in test_mode,
the annotation will be save in `eval_ann_infos`, which can be used
in Evaluator. Defaults to True.
file_client_args (dict): Configuration of file client.
Defaults to `dict(backend='disk')`.
Defaults to dict(backend='disk').
"""
METAINFO = {
'CLASSES': None, # names of all classes data used for the task
'PALETTE': None, # official color for visualization
'classes': None, # names of all classes data used for the task
'palette': None, # official color for visualization
'seg_valid_class_ids': None, # class_ids used for training
'seg_all_class_ids': None, # all possible class_ids in loaded seg mask
}
......@@ -62,12 +67,13 @@ class Seg3DDataset(BaseDataset):
pts='points',
img='',
pts_instance_mask='',
pts_emantic_mask=''),
pts_semantic_mask=''),
pipeline: List[Union[dict, Callable]] = [],
modality: dict = dict(use_lidar=True, use_camera=False),
ignore_index: Optional[int] = None,
scene_idxs: Optional[str] = None,
scene_idxs: Optional[Union[str, np.ndarray]] = None,
test_mode: bool = False,
serialize_data: bool = False,
load_eval_anns: bool = True,
file_client_args: dict = dict(backend='disk'),
**kwargs) -> None:
......@@ -78,11 +84,11 @@ class Seg3DDataset(BaseDataset):
# TODO: We maintain the ignore_index attributes,
# but we may consider to remove it in the future.
self.ignore_index = len(self.METAINFO['CLASSES']) if \
self.ignore_index = len(self.METAINFO['classes']) if \
ignore_index is None else ignore_index
# Get label mapping for custom classes
new_classes = metainfo.get('CLASSES', None)
new_classes = metainfo.get('classes', None)
self.label_mapping, self.label2cat, seg_valid_class_ids = \
self.get_label_mapping(new_classes)
......@@ -95,10 +101,10 @@ class Seg3DDataset(BaseDataset):
# generate palette if it is not defined based on
# label mapping, otherwise directly use palette
# defined in dataset config.
palette = metainfo.get('PALETTE', None)
palette = metainfo.get('palette', None)
updated_palette = self._update_palette(new_classes, palette)
metainfo['PALETTE'] = updated_palette
metainfo['palette'] = updated_palette
# construct seg_label_mapping for semantic mask
seg_max_cat_id = len(self.METAINFO['seg_all_class_ids'])
......@@ -117,18 +123,19 @@ class Seg3DDataset(BaseDataset):
data_prefix=data_prefix,
pipeline=pipeline,
test_mode=test_mode,
serialize_data=serialize_data,
**kwargs)
self.metainfo['seg_label_mapping'] = self.seg_label_mapping
self.scene_idxs = self.get_scene_idxs(scene_idxs)
self.data_list = [self.data_list[i] for i in self.scene_idxs]
# set group flag for the sampler
if not self.test_mode:
self._set_group_flag()
def get_label_mapping(self,
new_classes: Optional[Sequence] = None
) -> Union[Dict, None]:
new_classes: Optional[Sequence] = None) -> tuple:
"""Get label mapping.
The ``label_mapping`` is a dictionary, its keys are the old label ids
......@@ -138,21 +145,20 @@ class Seg3DDataset(BaseDataset):
None, `label_mapping` is not None.
Args:
new_classes (list, tuple, optional): The new classes name from
metainfo. Default to None.
new_classes (list or tuple, optional): The new classes name from
metainfo. Defaults to None.
Returns:
tuple: The mapping from old classes in cls.METAINFO to
new classes in metainfo
"""
old_classes = self.METAINFO.get('CLASSES', None)
old_classes = self.METAINFO.get('classes', None)
if (new_classes is not None and old_classes is not None
and list(new_classes) != list(old_classes)):
if not set(new_classes).issubset(old_classes):
raise ValueError(
f'new classes {new_classes} is not a '
f'subset of CLASSES {old_classes} in METAINFO.')
f'subset of classes {old_classes} in METAINFO.')
# obtain true id from valid_class_ids
valid_class_ids = [
......@@ -180,13 +186,14 @@ class Seg3DDataset(BaseDataset):
# map label to category name
label2cat = {
i: cat_name
for i, cat_name in enumerate(self.METAINFO['CLASSES'])
for i, cat_name in enumerate(self.METAINFO['classes'])
}
valid_class_ids = self.METAINFO['seg_valid_class_ids']
return label_mapping, label2cat, valid_class_ids
def _update_palette(self, new_classes, palette) -> list:
def _update_palette(self, new_classes: list, palette: Union[None,
list]) -> list:
"""Update palette according to metainfo.
If length of palette is equal to classes, just return the palette.
......@@ -199,10 +206,10 @@ class Seg3DDataset(BaseDataset):
"""
if palette is None:
# If palette is not defined, it generate a palette according
# to the original PALETTE and classes.
old_classes = self.METAINFO.get('CLASSES', None)
# to the original palette and classes.
old_classes = self.METAINFO.get('classes', None)
palette = [
self.METAINFO['PALETTE'][old_classes.index(cls_name)]
self.METAINFO['palette'][old_classes.index(cls_name)]
for cls_name in new_classes
]
return palette
......@@ -211,8 +218,8 @@ class Seg3DDataset(BaseDataset):
if len(palette) == len(new_classes):
return palette
else:
raise ValueError('Once PLATTE in set in metainfo, it should'
'match CLASSES in metainfo')
raise ValueError('Once palette in set in metainfo, it should'
'match classes in metainfo')
def parse_data_info(self, info: dict) -> dict:
"""Process the raw data info.
......@@ -260,7 +267,8 @@ class Seg3DDataset(BaseDataset):
return info
def get_scene_idxs(self, scene_idxs):
def get_scene_idxs(self, scene_idxs: Union[None, str,
np.ndarray]) -> np.ndarray:
"""Compute scene_idxs for data sampling.
We sample more times for scenes with more points.
......@@ -282,7 +290,7 @@ class Seg3DDataset(BaseDataset):
return scene_idxs.astype(np.int32)
def _set_group_flag(self):
def _set_group_flag(self) -> None:
"""Set flag according to image aspect ratio.
Images with aspect ratio greater than 1 will be set as group 1,
......
mmdet3d/datasets/semantickitti_dataset.py
View file @ d7067e44
# Copyright (c) OpenMMLab. All rights reserved.
from typing import Callable, List, Optional, Union
import numpy as np
from mmdet3d.registry import DATASETS
from .seg3d_dataset import Seg3DDataset
......@@ -14,30 +16,35 @@ class SemanticKITTIDataset(Seg3DDataset):
for data downloading
Args:
data_root (str): Path of dataset root.
ann_file (str): Path of annotation file.
pipeline (list[dict], optional): Pipeline used for data processing.
data_root (str, optional): Path of dataset root. Defaults to None.
ann_file (str): Path of annotation file. Defaults to ''.
metainfo (dict, optional): Meta information for dataset, such as class
information. Defaults to None.
data_prefix (dict): Prefix for training data. Defaults to
dict(pts='points',
img='',
pts_instance_mask='',
pts_semantic_mask='').
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used as input,
it usually has following keys:
- use_camera: bool
- use_lidar: bool
Defaults to dict(use_lidar=True, use_camera=False).
ignore_index (int, optional): The label index to be ignored, e.g.
unannotated points. If None is given, set to len(self.classes) to
be consistent with PointSegClassMapping function in pipeline.
Defaults to None.
classes (tuple[str], optional): Classes used in the dataset.
scene_idxs (np.ndarray or str, optional): Precomputed index to load
data. For scenes with many points, we may sample it several times.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input. Defaults to None.
box_type_3d (str, optional): NO 3D box for this dataset.
You can choose any type
Based on the `box_type_3d`, the dataset will encapsulate the box
to its original format then converted them to `box_type_3d`.
Defaults to 'LiDAR' in this dataset. Available options includes
- 'LiDAR': Box in LiDAR coordinates.
- 'Depth': Box in depth coordinates, usually for indoor dataset.
- 'Camera': Box in camera coordinates.
filter_empty_gt (bool, optional): Whether to filter empty GT.
Defaults to True.
test_mode (bool, optional): Whether the dataset is in test mode.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
"""
METAINFO = {
'CLASSES': ('unlabeled', 'car', 'bicycle', 'motorcycle', 'truck',
'classes': ('unlabeled', 'car', 'bicycle', 'motorcycle', 'truck',
'bus', 'person', 'bicyclist', 'motorcyclist', 'road',
'parking', 'sidewalk', 'other-ground', 'building', 'fence',
'vegetation', 'trunck', 'terrian', 'pole', 'traffic-sign'),
......@@ -52,12 +59,15 @@ class SemanticKITTIDataset(Seg3DDataset):
ann_file: str = '',
metainfo: Optional[dict] = None,
data_prefix: dict = dict(
pts='points', img='', instance_mask='', semantic_mask=''),
pts='points',
img='',
pts_instance_mask='',
pts_semantic_mask=''),
pipeline: List[Union[dict, Callable]] = [],
modality: dict = dict(use_lidar=True, use_camera=False),
ignore_index=None,
scene_idxs=None,
test_mode=False,
ignore_index: Optional[int] = None,
scene_idxs: Optional[Union[str, np.ndarray]] = None,
test_mode: bool = False,
**kwargs) -> None:
super().__init__(
......
mmdet3d/datasets/sunrgbd_dataset.py
View file @ d7067e44
......@@ -24,13 +24,13 @@ class SUNRGBDDataset(Det3DDataset):
ann_file (str): Path of annotation file.
metainfo (dict, optional): Meta information for dataset, such as class
information. Defaults to None.
data_prefix (dict, optiona;): Prefix for data. Defaults to
data_prefix (dict): Prefix for data. Defaults to
dict(pts='points',img='sunrgbd_trainval').
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
as input. Defaults to dict(use_camera=True, use_lidar=True).
default_cam_key (str, optional): The default camera name adopted.
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used as input.
Defaults to dict(use_camera=True, use_lidar=True).
default_cam_key (str): The default camera name adopted.
Defaults to 'CAM0'.
box_type_3d (str): Type of 3D box of this dataset.
Based on the `box_type_3d`, the dataset will encapsulate the box
......@@ -40,13 +40,13 @@ class SUNRGBDDataset(Det3DDataset):
- 'LiDAR': Box in LiDAR coordinates.
- 'Depth': Box in depth coordinates, usually for indoor dataset.
- 'Camera': Box in camera coordinates.
filter_empty_gt (bool, optional): Whether to filter empty GT.
filter_empty_gt (bool): Whether to filter empty GT.
Defaults to True.
test_mode (bool, optional): Whether the dataset is in test mode.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
"""
METAINFO = {
'CLASSES': ('bed', 'table', 'sofa', 'chair', 'toilet', 'desk',
'classes': ('bed', 'table', 'sofa', 'chair', 'toilet', 'desk',
'dresser', 'night_stand', 'bookshelf', 'bathtub')
}
......@@ -58,11 +58,11 @@ class SUNRGBDDataset(Det3DDataset):
pts='points', img='sunrgbd_trainval/image'),
pipeline: List[Union[dict, Callable]] = [],
default_cam_key: str = 'CAM0',
modality=dict(use_camera=True, use_lidar=True),
modality: dict = dict(use_camera=True, use_lidar=True),
box_type_3d: str = 'Depth',
filter_empty_gt: bool = True,
test_mode: bool = False,
**kwargs):
**kwargs) -> None:
super().__init__(
data_root=data_root,
ann_file=ann_file,
......@@ -121,7 +121,7 @@ class SUNRGBDDataset(Det3DDataset):
return info
def parse_ann_info(self, info: dict) -> dict:
"""Process the `instances` in data info to `ann_info`
"""Process the `instances` in data info to `ann_info`.
Args:
info (dict): Info dict.
......
mmdet3d/datasets/transforms/__init__.py
View file @ d7067e44
# Copyright (c) OpenMMLab. All rights reserved.
from .compose import Compose
from .dbsampler import DataBaseSampler
from .formating import Pack3DDetInputs
from .loading import (LoadAnnotations3D, LoadImageFromFileMono3D,
......@@ -21,15 +20,13 @@ from .transforms_3d import (AffineResize, BackgroundPointsFilter,
__all__ = [
'ObjectSample', 'RandomFlip3D', 'ObjectNoise', 'GlobalRotScaleTrans',
'PointShuffle', 'ObjectRangeFilter', 'PointsRangeFilter',
'Pack3DDetInputs',
'Compose', 'LoadMultiViewImageFromFiles', 'LoadPointsFromFile',
'DataBaseSampler',
'NormalizePointsColor', 'LoadAnnotations3D', 'IndoorPointSample',
'PointSample', 'PointSegClassMapping', 'MultiScaleFlipAug3D',
'LoadPointsFromMultiSweeps', 'BackgroundPointsFilter',
'VoxelBasedPointSampler', 'GlobalAlignment', 'IndoorPatchPointSample',
'LoadImageFromFileMono3D', 'ObjectNameFilter', 'RandomDropPointsColor',
'RandomJitterPoints', 'AffineResize', 'RandomShiftScale',
'LoadPointsFromDict', 'Resize3D', 'RandomResize3D',
'Pack3DDetInputs', 'LoadMultiViewImageFromFiles', 'LoadPointsFromFile',
'DataBaseSampler', 'NormalizePointsColor', 'LoadAnnotations3D',
'IndoorPointSample', 'PointSample', 'PointSegClassMapping',
'MultiScaleFlipAug3D', 'LoadPointsFromMultiSweeps',
'BackgroundPointsFilter', 'VoxelBasedPointSampler', 'GlobalAlignment',
'IndoorPatchPointSample', 'LoadImageFromFileMono3D', 'ObjectNameFilter',
'RandomDropPointsColor', 'RandomJitterPoints', 'AffineResize',
'RandomShiftScale', 'LoadPointsFromDict', 'Resize3D', 'RandomResize3D',
'MultiViewWrapper', 'PhotoMetricDistortion3D'
]
mmdet3d/datasets/transforms/compose.py deleted 100644 → 0
View file @ 28fe73d2
# Copyright (c) OpenMMLab. All rights reserved.
import collections
from mmdet3d.registry import TRANSFORMS
@TRANSFORMS.register_module()
class Compose:
"""Compose multiple transforms sequentially.
Args:
transforms (Sequence[dict | callable]): Sequence of transform object or
config dict to be composed.
"""
def __init__(self, transforms):
assert isinstance(transforms, collections.abc.Sequence)
self.transforms = []
for transform in transforms:
if isinstance(transform, dict):
transform = TRANSFORMS.build(transform)
self.transforms.append(transform)
elif callable(transform):
self.transforms.append(transform)
else:
raise TypeError('transform must be callable or a dict')
def __call__(self, data):
"""Call function to apply transforms sequentially.
Args:
data (dict): A result dict contains the data to transform.
Returns:
dict: Transformed data.
"""
for t in self.transforms:
data = t(data)
if data is None:
return None
return data
def __repr__(self):
format_string = self.__class__.__name__ + '('
for t in self.transforms:
str_ = t.__repr__()
if 'Compose(' in str_:
str_ = str_.replace('\n', '\n ')
format_string += '\n'
format_string += f' {str_}'
format_string += '\n)'
return format_string
mmdet3d/datasets/transforms/dbsampler.py
View file @ d7067e44
......@@ -18,9 +18,8 @@ class BatchSampler:
sample_list (list[dict]): List of samples.
name (str, optional): The category of samples. Defaults to None.
epoch (int, optional): Sampling epoch. Defaults to None.
shuffle (bool, optional): Whether to shuffle indices.
Defaults to False.
drop_reminder (bool, optional): Drop reminder. Defaults to False.
shuffle (bool): Whether to shuffle indices. Defaults to False.
drop_reminder (bool): Drop reminder. Defaults to False.
"""
def __init__(self,
......@@ -90,12 +89,11 @@ class DataBaseSampler(object):
prepare (dict): Name of preparation functions and the input value.
sample_groups (dict): Sampled classes and numbers.
classes (list[str], optional): List of classes. Defaults to None.
points_loader(dict, optional): Config of points loader. Defaults to
points_loader (dict): Config of points loader. Defaults to
dict(type='LoadPointsFromFile', load_dim=4, use_dim=[0, 1, 2, 3]).
file_client_args (dict, optional): Config dict of file clients,
refer to
https://github.com/open-mmlab/mmengine/blob/main/mmengine/fileio/file_client.py
for more details. Defaults to dict(backend='disk').
file_client_args (dict): Arguments to instantiate a FileClient.
See :class:`mmengine.fileio.FileClient` for details.
Defaults to dict(backend='disk').
"""
def __init__(
......@@ -133,12 +131,12 @@ class DataBaseSampler(object):
from mmengine.logging import MMLogger
logger: MMLogger = MMLogger.get_current_instance()
for k, v in db_infos.items():
logger.info(f'load {len(v)} {k} database infos')
logger.info(f'load {len(v)} {k} database infos in DataBaseSampler')
for prep_func, val in prepare.items():
db_infos = getattr(self, prep_func)(db_infos, val)
logger.info('After filter database:')
for k, v in db_infos.items():
logger.info(f'load {len(v)} {k} database infos')
logger.info(f'load {len(v)} {k} database infos in DataBaseSampler')
self.db_infos = db_infos
......
mmdet3d/datasets/transforms/formating.py
View file @ d7067e44
......@@ -102,7 +102,7 @@ class Pack3DDetInputs(BaseTransform):
- points
- img
- 'data_samples' (obj:`Det3DDataSample`): The annotation info of
- 'data_samples' (:obj:`Det3DDataSample`): The annotation info of
the sample.
"""
# augtest
......
mmdet3d/datasets/transforms/loading.py
View file @ d7067e44
......@@ -7,10 +7,10 @@ import mmengine
import numpy as np
from mmcv.transforms import LoadImageFromFile
from mmcv.transforms.base import BaseTransform
from mmdet.datasets.transforms import LoadAnnotations
from mmdet3d.registry import TRANSFORMS
from mmdet3d.structures.points import BasePoints, get_points_type
from mmdet.datasets.transforms import LoadAnnotations
@TRANSFORMS.register_module()
......@@ -20,19 +20,17 @@ class LoadMultiViewImageFromFiles(BaseTransform):
Expects results['img_filename'] to be a list of filenames.
Args:
to_float32 (bool, optional): Whether to convert the img to float32.
to_float32 (bool): Whether to convert the img to float32.
Defaults to False.
color_type (str, optional): Color type of the file.
Defaults to 'unchanged'.
file_client_args (dict): Config dict of file clients,
refer to
https://github.com/open-mmlab/mmcv/blob/master/mmcv/fileio/file_client.py
for more details. Defaults to dict(backend='disk').
num_views (int): num of view in a frame. Default to 5.
num_ref_frames (int): num of frame in loading. Default to -1.
test_mode (bool): Whether is test mode in loading. Default to False.
set_default_scale (bool): Whether to set default scale. Default to
True.
color_type (str): Color type of the file. Defaults to 'unchanged'.
file_client_args (dict): Arguments to instantiate a FileClient.
See :class:`mmengine.fileio.FileClient` for details.
Defaults to dict(backend='disk').
num_views (int): Number of view in a frame. Defaults to 5.
num_ref_frames (int): Number of frame in loading. Defaults to -1.
test_mode (bool): Whether is test mode in loading. Defaults to False.
set_default_scale (bool): Whether to set default scale.
Defaults to True.
"""
def __init__(self,
......@@ -210,7 +208,7 @@ class LoadMultiViewImageFromFiles(BaseTransform):
results['num_ref_frames'] = self.num_ref_frames
return results
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(to_float32={self.to_float32}, '
......@@ -276,22 +274,17 @@ class LoadPointsFromMultiSweeps(BaseTransform):
This is usually used for nuScenes dataset to utilize previous sweeps.
Args:
sweeps_num (int, optional): Number of sweeps. Defaults to 10.
load_dim (int, optional): Dimension number of the loaded points.
Defaults to 5.
use_dim (list[int], optional): Which dimension to use.
Defaults to [0, 1, 2, 4].
file_client_args (dict, optional): Config dict of file clients,
refer to
https://github.com/open-mmlab/mmengine/blob/main/mmengine/fileio/file_client.py
for more details. Defaults to dict(backend='disk').
pad_empty_sweeps (bool, optional): Whether to repeat keyframe when
sweeps_num (int): Number of sweeps. Defaults to 10.
load_dim (int): Dimension number of the loaded points. Defaults to 5.
use_dim (list[int]): Which dimension to use. Defaults to [0, 1, 2, 4].
file_client_args (dict): Arguments to instantiate a FileClient.
See :class:`mmengine.fileio.FileClient` for details.
Defaults to dict(backend='disk').
pad_empty_sweeps (bool): Whether to repeat keyframe when
sweeps is empty. Defaults to False.
remove_close (bool, optional): Whether to remove close points.
Defaults to False.
test_mode (bool, optional): If `test_mode=True`, it will not
randomly sample sweeps but select the nearest N frames.
Defaults to False.
remove_close (bool): Whether to remove close points. Defaults to False.
test_mode (bool): If `test_mode=True`, it will not randomly sample
sweeps but select the nearest N frames. Defaults to False.
"""
def __init__(self,
......@@ -336,11 +329,11 @@ class LoadPointsFromMultiSweeps(BaseTransform):
def _remove_close(self,
points: Union[np.ndarray, BasePoints],
radius: float = 1.0) -> Union[np.ndarray, BasePoints]:
"""Removes point too close within a certain radius from origin.
"""Remove point too close within a certain radius from origin.
Args:
points (np.ndarray | :obj:`BasePoints`): Sweep points.
radius (float, optional): Radius below which points are removed.
radius (float): Radius below which points are removed.
Defaults to 1.0.
Returns:
......@@ -414,7 +407,7 @@ class LoadPointsFromMultiSweeps(BaseTransform):
results['points'] = points
return results
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
return f'{self.__class__.__name__}(sweeps_num={self.sweeps_num})'
......@@ -465,7 +458,7 @@ class PointSegClassMapping(BaseTransform):
return results
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
return repr_str
......@@ -505,7 +498,7 @@ class NormalizePointsColor(BaseTransform):
input_dict['points'] = points
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(color_mean={self.color_mean})'
......@@ -533,19 +526,15 @@ class LoadPointsFromFile(BaseTransform):
- 'LIDAR': Points in LiDAR coordinates.
- 'DEPTH': Points in depth coordinates, usually for indoor dataset.
- 'CAMERA': Points in camera coordinates.
load_dim (int, optional): The dimension of the loaded points.
Defaults to 6.
use_dim (list[int] | int, optional): Which dimensions of the points
to use. Defaults to [0, 1, 2]. For KITTI dataset, set use_dim=4
load_dim (int): The dimension of the loaded points. Defaults to 6.
use_dim (list[int] | int): Which dimensions of the points to use.
Defaults to [0, 1, 2]. For KITTI dataset, set use_dim=4
or use_dim=[0, 1, 2, 3] to use the intensity dimension.
shift_height (bool, optional): Whether to use shifted height.
Defaults to False.
use_color (bool, optional): Whether to use color features.
Defaults to False.
file_client_args (dict, optional): Config dict of file clients,
refer to
https://github.com/open-mmlab/mmengine/blob/main/mmengine/fileio/file_client.py
for more details. Defaults to dict(backend='disk').
shift_height (bool): Whether to use shifted height. Defaults to False.
use_color (bool): Whether to use color features. Defaults to False.
file_client_args (dict): Arguments to instantiate a FileClient.
See :class:`mmengine.fileio.FileClient` for details.
Defaults to dict(backend='disk').
"""
def __init__(
......@@ -638,7 +627,7 @@ class LoadPointsFromFile(BaseTransform):
return results
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__ + '('
repr_str += f'shift_height={self.shift_height}, '
......@@ -688,7 +677,7 @@ class LoadAnnotations3D(LoadAnnotations):
- pts_instance_mask_path (str): Path of instance mask file.
Only when `with_mask_3d` is True.
- pts_semantic_mask_path (str): Path of semantic mask file.
Only when
Only when `with_seg_3d` is True.
Added Keys:
......@@ -713,33 +702,25 @@ class LoadAnnotations3D(LoadAnnotations):
Only when `with_seg_3d` is True.
Args:
with_bbox_3d (bool, optional): Whether to load 3D boxes.
Defaults to True.
with_label_3d (bool, optional): Whether to load 3D labels.
Defaults to True.
with_attr_label (bool, optional): Whether to load attribute label.
Defaults to False.
with_mask_3d (bool, optional): Whether to load 3D instance masks.
for points. Defaults to False.
with_seg_3d (bool, optional): Whether to load 3D semantic masks.
for points. Defaults to False.
with_bbox (bool, optional): Whether to load 2D boxes.
Defaults to False.
with_label (bool, optional): Whether to load 2D labels.
with_bbox_3d (bool): Whether to load 3D boxes. Defaults to True.
with_label_3d (bool): Whether to load 3D labels. Defaults to True.
with_attr_label (bool): Whether to load attribute label.
Defaults to False.
with_mask (bool, optional): Whether to load 2D instance masks.
with_mask_3d (bool): Whether to load 3D instance masks for points.
Defaults to False.
with_seg (bool, optional): Whether to load 2D semantic masks.
with_seg_3d (bool): Whether to load 3D semantic masks for points.
Defaults to False.
with_bbox_depth (bool, optional): Whether to load 2.5D boxes.
Defaults to False.
poly2mask (bool, optional): Whether to convert polygon annotations
to bitmasks. Defaults to True.
seg_3d_dtype (dtype, optional): Dtype of 3D semantic masks.
Defaults to int64.
file_client_args (dict): Config dict of file clients, refer to
https://github.com/open-mmlab/mmengine/blob/main/mmengine/fileio/file_client.py
for more details.
with_bbox (bool): Whether to load 2D boxes. Defaults to False.
with_label (bool): Whether to load 2D labels. Defaults to False.
with_mask (bool): Whether to load 2D instance masks. Defaults to False.
with_seg (bool): Whether to load 2D semantic masks. Defaults to False.
with_bbox_depth (bool): Whether to load 2.5D boxes. Defaults to False.
poly2mask (bool): Whether to convert polygon annotations to bitmasks.
Defaults to True.
seg_3d_dtype (dtype): Dtype of 3D semantic masks. Defaults to int64.
file_client_args (dict): Arguments to instantiate a FileClient.
See :class:`mmengine.fileio.FileClient` for details.
Defaults to dict(backend='disk').
"""
def __init__(
......@@ -889,7 +870,8 @@ class LoadAnnotations3D(LoadAnnotations):
`ignore_flag`
Args:
results (dict): Result dict from :obj:``mmcv.BaseDataset``.
results (dict): Result dict from :obj:`mmcv.BaseDataset`.
Returns:
dict: The dict contains loaded bounding box annotations.
"""
......@@ -900,7 +882,7 @@ class LoadAnnotations3D(LoadAnnotations):
"""Private function to load label annotations.
Args:
results (dict): Result dict from :obj :obj:``mmcv.BaseDataset``.
results (dict): Result dict from :obj :obj:`mmcv.BaseDataset`.
Returns:
dict: The dict contains loaded label annotations.
......@@ -933,7 +915,7 @@ class LoadAnnotations3D(LoadAnnotations):
return results
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
indent_str = ' '
repr_str = self.__class__.__name__ + '(\n'
......
mmdet3d/datasets/transforms/test_time_aug.py
View file @ d7067e44
......@@ -19,18 +19,17 @@ class MultiScaleFlipAug3D(BaseTransform):
img_scale (tuple | list[tuple]): Images scales for resizing.
pts_scale_ratio (float | list[float]): Points scale ratios for
resizing.
flip (bool, optional): Whether apply flip augmentation.
Defaults to False.
flip_direction (str | list[str], optional): Flip augmentation
directions for images, options are "horizontal" and "vertical".
flip (bool): Whether apply flip augmentation. Defaults to False.
flip_direction (str | list[str]): Flip augmentation directions
for images, options are "horizontal" and "vertical".
If flip_direction is list, multiple flip augmentations will
be applied. It has no effect when ``flip == False``.
Defaults to 'horizontal'.
pcd_horizontal_flip (bool, optional): Whether to apply horizontal
flip augmentation to point cloud. Defaults to True.
pcd_horizontal_flip (bool): Whether to apply horizontal flip
augmentation to point cloud. Defaults to False.
Note that it works only when 'flip' is turned on.
pcd_vertical_flip (bool, optional): Whether to apply vertical flip
augmentation to point cloud. Defaults to True.
pcd_vertical_flip (bool): Whether to apply vertical flip
augmentation to point cloud. Defaults to False.
Note that it works only when 'flip' is turned on.
"""
......@@ -112,7 +111,7 @@ class MultiScaleFlipAug3D(BaseTransform):
return aug_data_list
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(transforms={self.transforms}, '
......
mmdet3d/datasets/transforms/transforms_3d.py
View file @ d7067e44
......@@ -6,7 +6,9 @@ from typing import List, Optional, Tuple, Union
import cv2
import mmcv
import numpy as np
from mmcv.transforms import BaseTransform, RandomResize, Resize
from mmcv.transforms import BaseTransform, Compose, RandomResize, Resize
from mmdet.datasets.transforms import (PhotoMetricDistortion, RandomCrop,
RandomFlip)
from mmengine import is_tuple_of
from mmdet3d.models.task_modules import VoxelGenerator
......@@ -15,9 +17,6 @@ from mmdet3d.structures import (CameraInstance3DBoxes, DepthInstance3DBoxes,
LiDARInstance3DBoxes)
from mmdet3d.structures.ops import box_np_ops
from mmdet3d.structures.points import BasePoints
from mmdet.datasets.transforms import (PhotoMetricDistortion, RandomCrop,
RandomFlip)
from .compose import Compose
from .data_augment_utils import noise_per_object_v3_
......@@ -30,7 +29,7 @@ class RandomDropPointsColor(BaseTransform):
util/transform.py#L223>`_ for more details.
Args:
drop_ratio (float, optional): The probability of dropping point colors.
drop_ratio (float): The probability of dropping point colors.
Defaults to 0.2.
"""
......@@ -46,8 +45,8 @@ class RandomDropPointsColor(BaseTransform):
input_dict (dict): Result dict from loading pipeline.
Returns:
dict: Results after color dropping,
'points' key is updated in the result dict.
dict: Results after color dropping, 'points' key is updated
in the result dict.
"""
points = input_dict['points']
assert points.attribute_dims is not None and \
......@@ -64,7 +63,7 @@ class RandomDropPointsColor(BaseTransform):
points.color = points.color * 0.0
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(drop_ratio={self.drop_ratio})'
......@@ -108,8 +107,8 @@ class RandomFlip3D(RandomFlip):
in vertical direction. Defaults to 0.0.
flip_box3d (bool): Whether to flip bounding box. In most of the case,
the box should be fliped. In cam-based bev detection, this is set
to false, since the flip of 2D images does not influence the 3D
box. Default to True.
to False, since the flip of 2D images does not influence the 3D
box. Defaults to True.
"""
def __init__(self,
......@@ -150,8 +149,7 @@ class RandomFlip3D(RandomFlip):
Args:
input_dict (dict): Result dict from loading pipeline.
direction (str, optional): Flip direction.
Default: 'horizontal'.
direction (str): Flip direction. Defaults to 'horizontal'.
Returns:
dict: Flipped results, 'points', 'bbox3d_fields' keys are
......@@ -241,7 +239,7 @@ class RandomFlip3D(RandomFlip):
input_dict['transformation_3d_flow'].extend(['VF'])
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(sync_2d={self.sync_2d},'
......@@ -309,7 +307,7 @@ class RandomJitterPoints(BaseTransform):
points.translate(jitter_noise)
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(jitter_std={self.jitter_std},'
......@@ -344,11 +342,11 @@ class ObjectSample(BaseTransform):
Args:
db_sampler (dict): Config dict of the database sampler.
sample_2d (bool): Whether to also paste 2D image patch to the images
sample_2d (bool): Whether to also paste 2D image patch to the images.
This should be true when applying multi-modality cut-and-paste.
Defaults to False.
use_ground_plane (bool): Whether to use ground plane to adjust the
3D labels.
3D labels. Defaults to False.
"""
def __init__(self,
......@@ -445,12 +443,12 @@ class ObjectSample(BaseTransform):
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'db_sampler={self.db_sampler},'
repr_str += f'(db_sampler={self.db_sampler},'
repr_str += f' sample_2d={self.sample_2d},'
repr_str += f' use_ground_plane={self.use_ground_plane}'
repr_str += f' use_ground_plane={self.use_ground_plane})'
return repr_str
......@@ -469,15 +467,15 @@ class ObjectNoise(BaseTransform):
- gt_bboxes_3d
Args:
translation_std (list[float], optional): Standard deviation of the
translation_std (list[float]): Standard deviation of the
distribution where translation noise are sampled from.
Defaults to [0.25, 0.25, 0.25].
global_rot_range (list[float], optional): Global rotation to the scene.
global_rot_range (list[float]): Global rotation to the scene.
Defaults to [0.0, 0.0].
rot_range (list[float], optional): Object rotation range.
rot_range (list[float]): Object rotation range.
Defaults to [-0.15707963267, 0.15707963267].
num_try (int, optional): Number of times to try if the noise applied is
invalid. Defaults to 100.
num_try (int): Number of times to try if the noise applied is invalid.
Defaults to 100.
"""
def __init__(self,
......@@ -519,7 +517,7 @@ class ObjectNoise(BaseTransform):
input_dict['points'] = points.new_point(numpy_points)
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(num_try={self.num_try},'
......@@ -610,7 +608,7 @@ class GlobalAlignment(BaseTransform):
return results
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(rotation_axis={self.rotation_axis})'
......@@ -640,15 +638,15 @@ class GlobalRotScaleTrans(BaseTransform):
- pcd_scale_factor (np.float32)
Args:
rot_range (list[float], optional): Range of rotation angle.
rot_range (list[float]): Range of rotation angle.
Defaults to [-0.78539816, 0.78539816] (close to [-pi/4, pi/4]).
scale_ratio_range (list[float], optional): Range of scale ratio.
scale_ratio_range (list[float]): Range of scale ratio.
Defaults to [0.95, 1.05].
translation_std (list[float], optional): The standard deviation of
translation_std (list[float]): The standard deviation of
translation noise applied to a scene, which
is sampled from a gaussian distribution whose standard deviation
is set by ``translation_std``. Defaults to [0, 0, 0]
shift_height (bool, optional): Whether to shift height.
is set by ``translation_std``. Defaults to [0, 0, 0].
shift_height (bool): Whether to shift height.
(the fourth dimension of indoor points) when scaling.
Defaults to False.
"""
......@@ -689,8 +687,7 @@ class GlobalRotScaleTrans(BaseTransform):
Returns:
dict: Results after translation, 'points', 'pcd_trans'
and `gt_bboxes_3d` is updated
in the result dict.
and `gt_bboxes_3d` is updated in the result dict.
"""
translation_std = np.array(self.translation_std, dtype=np.float32)
trans_factor = np.random.normal(scale=translation_std, size=3).T
......@@ -708,8 +705,7 @@ class GlobalRotScaleTrans(BaseTransform):
Returns:
dict: Results after rotation, 'points', 'pcd_rotation'
and `gt_bboxes_3d` is updated
in the result dict.
and `gt_bboxes_3d` is updated in the result dict.
"""
rotation = self.rot_range
noise_rotation = np.random.uniform(rotation[0], rotation[1])
......@@ -735,8 +731,7 @@ class GlobalRotScaleTrans(BaseTransform):
Returns:
dict: Results after scaling, 'points' and
`gt_bboxes_3d` is updated
in the result dict.
`gt_bboxes_3d` is updated in the result dict.
"""
scale = input_dict['pcd_scale_factor']
points = input_dict['points']
......@@ -774,7 +769,7 @@ class GlobalRotScaleTrans(BaseTransform):
Returns:
dict: Results after scaling, 'points', 'pcd_rotation',
'pcd_scale_factor', 'pcd_trans' and `gt_bboxes_3d` is updated
'pcd_scale_factor', 'pcd_trans' and `gt_bboxes_3d` are updated
in the result dict.
"""
if 'transformation_3d_flow' not in input_dict:
......@@ -791,7 +786,7 @@ class GlobalRotScaleTrans(BaseTransform):
input_dict['transformation_3d_flow'].extend(['R', 'S', 'T'])
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(rot_range={self.rot_range},'
......@@ -829,7 +824,7 @@ class PointShuffle(BaseTransform):
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
return self.__class__.__name__
......@@ -850,7 +845,7 @@ class ObjectRangeFilter(BaseTransform):
point_cloud_range (list[float]): Point cloud range.
"""
def __init__(self, point_cloud_range: List[float]):
def __init__(self, point_cloud_range: List[float]) -> None:
self.pcd_range = np.array(point_cloud_range, dtype=np.float32)
def transform(self, input_dict: dict) -> dict:
......@@ -887,7 +882,7 @@ class ObjectRangeFilter(BaseTransform):
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(point_cloud_range={self.pcd_range.tolist()})'
......@@ -942,7 +937,7 @@ class PointsRangeFilter(BaseTransform):
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(point_cloud_range={self.pcd_range.tolist()})'
......@@ -987,7 +982,7 @@ class ObjectNameFilter(BaseTransform):
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(classes={self.classes})'
......@@ -1017,8 +1012,8 @@ class PointSample(BaseTransform):
sample_range (float, optional): The range where to sample points.
If not None, the points with depth larger than `sample_range` are
prior to be sampled. Defaults to None.
replace (bool, optional): Whether the sampling is with or without
replacement. Defaults to False.
replace (bool): Whether the sampling is with or without replacement.
Defaults to False.
"""
def __init__(self,
......@@ -1046,10 +1041,9 @@ class PointSample(BaseTransform):
num_samples (int): Number of samples to be sampled.
sample_range (float, optional): Indicating the range where the
points will be sampled. Defaults to None.
replace (bool, optional): Sampling with or without replacement.
Defaults to False.
return_choices (bool, optional): Whether return choice.
replace (bool): Sampling with or without replacement.
Defaults to False.
return_choices (bool): Whether return choice. Defaults to False.
Returns:
tuple[:obj:`BasePoints`, np.ndarray] | :obj:`BasePoints`:
......@@ -1113,7 +1107,7 @@ class PointSample(BaseTransform):
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(num_points={self.num_points},'
......@@ -1149,7 +1143,7 @@ class IndoorPatchPointSample(BaseTransform):
Args:
num_points (int): Number of points to be sampled.
block_size (float, optional): Size of a block to sample points from.
block_size (float): Size of a block to sample points from.
Defaults to 1.5.
sample_rate (float, optional): Stride used in sliding patch generation.
This parameter is unused in `IndoorPatchPointSample` and thus has
......@@ -1159,17 +1153,17 @@ class IndoorPatchPointSample(BaseTransform):
segmentation task. This is set in PointSegClassMapping as neg_cls.
If not None, will be used as a patch selection criterion.
Defaults to None.
use_normalized_coord (bool, optional): Whether to use normalized xyz as
use_normalized_coord (bool): Whether to use normalized xyz as
additional features. Defaults to False.
num_try (int, optional): Number of times to try if the patch selected
is invalid. Defaults to 10.
enlarge_size (float, optional): Enlarge the sampled patch to
num_try (int): Number of times to try if the patch selected is invalid.
Defaults to 10.
enlarge_size (float): Enlarge the sampled patch to
[-block_size / 2 - enlarge_size, block_size / 2 + enlarge_size] as
an augmentation. If None, set it as 0. Defaults to 0.2.
min_unique_num (int, optional): Minimum number of unique points
the sampled patch should contain. If None, use PointNet++'s method
to judge uniqueness. Defaults to None.
eps (float, optional): A value added to patch boundary to guarantee
eps (float): A value added to patch boundary to guarantee
points coverage. Defaults to 1e-2.
Note:
......@@ -1386,7 +1380,7 @@ class IndoorPatchPointSample(BaseTransform):
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(num_points={self.num_points},'
......@@ -1405,7 +1399,7 @@ class BackgroundPointsFilter(BaseTransform):
"""Filter background points near the bounding box.
Args:
bbox_enlarge_range (tuple[float], float): Bbox enlarge range.
bbox_enlarge_range (tuple[float] | float): Bbox enlarge range.
"""
def __init__(self, bbox_enlarge_range: Union[Tuple[float], float]) -> None:
......@@ -1458,7 +1452,7 @@ class BackgroundPointsFilter(BaseTransform):
input_dict['pts_semantic_mask'] = pts_semantic_mask[valid_masks]
return input_dict
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(bbox_enlarge_range={self.bbox_enlarge_range.tolist()})'
......@@ -1473,9 +1467,10 @@ class VoxelBasedPointSampler(BaseTransform):
Args:
cur_sweep_cfg (dict): Config for sampling current points.
prev_sweep_cfg (dict): Config for sampling previous points.
prev_sweep_cfg (dict, optional): Config for sampling previous points.
Defaults to None.
time_dim (int): Index that indicate the time dimension
for input points.
for input points. Defaults to 3.
"""
def __init__(self,
......@@ -1502,7 +1497,7 @@ class VoxelBasedPointSampler(BaseTransform):
points (np.ndarray): Points subset to be sampled.
sampler (VoxelGenerator): Voxel based sampler for
each points subset.
point_dim (int): The dimension of each points
point_dim (int): The dimension of each points.
Returns:
np.ndarray: Sampled points.
......@@ -1589,7 +1584,7 @@ class VoxelBasedPointSampler(BaseTransform):
return results
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
def _auto_indent(repr_str, indent):
......@@ -1625,7 +1620,7 @@ class AffineResize(BaseTransform):
img_scale (tuple): Images scales for resizing.
down_ratio (int): The down ratio of feature map.
Actually the arg should be >= 1.
bbox_clip_border (bool, optional): Whether clip the objects
bbox_clip_border (bool): Whether clip the objects
outside the border of the image. Defaults to True.
"""
......@@ -1803,7 +1798,7 @@ class AffineResize(BaseTransform):
ref_point3 = ref_point2 + np.array([-d[1], d[0]])
return ref_point3
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(img_scale={self.img_scale}, '
......@@ -1863,7 +1858,7 @@ class RandomShiftScale(BaseTransform):
return results
def __repr__(self):
def __repr__(self) -> str:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(shift_scale={self.shift_scale}, '
......@@ -1874,7 +1869,7 @@ class RandomShiftScale(BaseTransform):
@TRANSFORMS.register_module()
class Resize3D(Resize):
def _resize_3d(self, results):
def _resize_3d(self, results: dict) -> None:
"""Resize centers_2d and modify camera intrinisc with
``results['scale']``."""
if 'centers_2d' in results:
......@@ -1888,6 +1883,7 @@ class Resize3D(Resize):
Args:
results (dict): Result dict from loading pipeline.
Returns:
dict: Resized results, 'img', 'gt_bboxes', 'gt_seg_map',
'gt_keypoints', 'scale', 'scale_factor', 'img_shape',
......@@ -1909,7 +1905,7 @@ class RandomResize3D(RandomResize):
and cam2img with ``results['scale']``.
"""
def _resize_3d(self, results):
def _resize_3d(self, results: dict) -> None:
"""Resize centers_2d and modify camera intrinisc with
``results['scale']``."""
if 'centers_2d' in results:
......@@ -1917,7 +1913,7 @@ class RandomResize3D(RandomResize):
results['cam2img'][0] *= np.array(results['scale_factor'][0])
results['cam2img'][1] *= np.array(results['scale_factor'][1])
def transform(self, results):
def transform(self, results: dict) -> dict:
"""Transform function to resize images, bounding boxes, masks, semantic
segmentation map. Compared to RandomResize, this function would further
check if scale is already set in results.
......@@ -1926,7 +1922,7 @@ class RandomResize3D(RandomResize):
results (dict): Result dict from loading pipeline.
Returns:
dict: Resized results, 'img_shape', 'pad_shape', 'scale_factor', \
dict: Resized results, 'img_shape', 'pad_shape', 'scale_factor',
'keep_ratio' keys are added into result dict.
"""
if 'scale' not in results:
......@@ -1989,9 +1985,9 @@ class RandomCrop3D(RandomCrop):
on cropped instance masks. Defaults to False.
bbox_clip_border (bool): Whether clip the objects outside
the border of the image. Defaults to True.
rel_offset_h (tuple): The cropping interval of image height. Default
rel_offset_h (tuple): The cropping interval of image height. Defaults
to (0., 1.).
rel_offset_w (tuple): The cropping interval of image width. Default
rel_offset_w (tuple): The cropping interval of image width. Defaults
to (0., 1.).
Note:
......@@ -2005,14 +2001,16 @@ class RandomCrop3D(RandomCrop):
``allow_negative_crop`` is set to False, skip this image.
"""
def __init__(self,
crop_size,
crop_type='absolute',
allow_negative_crop=False,
recompute_bbox=False,
bbox_clip_border=True,
rel_offset_h=(0., 1.),
rel_offset_w=(0., 1.)):
def __init__(
self,
crop_size: tuple,
crop_type: str = 'absolute',
allow_negative_crop: bool = False,
recompute_bbox: bool = False,
bbox_clip_border: bool = True,
rel_offset_h: tuple = (0., 1.),
rel_offset_w: tuple = (0., 1.)
) -> None:
super().__init__(
crop_size=crop_size,
crop_type=crop_type,
......@@ -2024,7 +2022,10 @@ class RandomCrop3D(RandomCrop):
self.rel_offset_h = rel_offset_h
self.rel_offset_w = rel_offset_w
def _crop_data(self, results, crop_size, allow_negative_crop):
def _crop_data(self,
results: dict,
crop_size: tuple,
allow_negative_crop: bool = False) -> dict:
"""Function to randomly crop images, bounding boxes, masks, semantic
segmentation maps.
......@@ -2032,7 +2033,7 @@ class RandomCrop3D(RandomCrop):
results (dict): Result dict from loading pipeline.
crop_size (tuple): Expected absolute size after cropping, (h, w).
allow_negative_crop (bool): Whether to allow a crop that does not
contain any bbox area. Default to False.
contain any bbox area. Defaults to False.
Returns:
dict: Randomly cropped results, 'img_shape' key in result dict is
......@@ -2119,7 +2120,7 @@ class RandomCrop3D(RandomCrop):
return results
def transform(self, results):
def transform(self, results: dict) -> dict:
"""Transform function to randomly crop images, bounding boxes, masks,
semantic segmentation maps.
......@@ -2139,7 +2140,8 @@ class RandomCrop3D(RandomCrop):
results = self._crop_data(results, crop_size, self.allow_negative_crop)
return results
def __repr__(self):
def __repr__(self) -> dict:
"""str: Return a string that describes the module."""
repr_str = self.__class__.__name__
repr_str += f'(crop_size={self.crop_size}, '
repr_str += f'crop_type={self.crop_type}, '
......@@ -2260,43 +2262,44 @@ class MultiViewWrapper(BaseTransform):
transforms (list[dict]): A list of dict specifying the transformations
for the monocular situation.
override_aug_config (bool): flag of whether to use the same aug config
for multiview image. Default to True.
for multiview image. Defaults to True.
process_fields (list): Desired keys that the transformations should
be conducted on. Default to ['img', 'cam2img', 'lidar2cam'],
be conducted on. Defaults to ['img', 'cam2img', 'lidar2cam'].
collected_keys (list): Collect information in transformation
like rotate angles, crop roi, and flip state. Default to
like rotate angles, crop roi, and flip state. Defaults to
['scale', 'scale_factor', 'crop',
'crop_offset', 'ori_shape',
'pad_shape', 'img_shape',
'pad_fixed_size', 'pad_size_divisor',
'flip', 'flip_direction', 'rotate'],
'flip', 'flip_direction', 'rotate'].
randomness_keys (list): The keys that related to the randomness
in transformation Default to
in transformation. Defaults to
['scale', 'scale_factor', 'crop_size', 'flip',
'flip_direction', 'photometric_param']
"""
def __init__(self,
def __init__(
self,
transforms: dict,
override_aug_config: bool = True,
process_fields: list = ['img', 'cam2img', 'lidar2cam'],
collected_keys: list = [
'scale', 'scale_factor', 'crop', 'img_crop_offset',
'ori_shape', 'pad_shape', 'img_shape', 'pad_fixed_size',
'pad_size_divisor', 'flip', 'flip_direction', 'rotate'
'scale', 'scale_factor', 'crop', 'img_crop_offset', 'ori_shape',
'pad_shape', 'img_shape', 'pad_fixed_size', 'pad_size_divisor',
'flip', 'flip_direction', 'rotate'
],
randomness_keys: list = [
'scale', 'scale_factor', 'crop_size', 'img_crop_offset',
'flip', 'flip_direction', 'photometric_param'
]):
'scale', 'scale_factor', 'crop_size', 'img_crop_offset', 'flip',
'flip_direction', 'photometric_param'
]
) -> None:
self.transforms = Compose(transforms)
self.override_aug_config = override_aug_config
self.collected_keys = collected_keys
self.process_fields = process_fields
self.randomness_keys = randomness_keys
def transform(self, input_dict):
def transform(self, input_dict: dict) -> dict:
"""Transform function to do the transform for multiview image.
Args:
......
mmdet3d/datasets/waymo_dataset.py
View file @ d7067e44
# Copyright (c) OpenMMLab. All rights reserved.
import os.path as osp
from typing import Callable, List, Optional, Union
from typing import Callable, List, Union
import numpy as np
......@@ -24,20 +24,20 @@ class WaymoDataset(KittiDataset):
data_root (str): Path of dataset root.
ann_file (str): Path of annotation file.
data_prefix (dict): data prefix for point cloud and
camera data dict. Default to dict(
camera data dict. Defaults to dict(
pts='velodyne',
CAM_FRONT='image_0',
CAM_FRONT_RIGHT='image_1',
CAM_FRONT_LEFT='image_2',
CAM_SIDE_RIGHT='image_3',
CAM_SIDE_LEFT='image_4')
pipeline (list[dict], optional): Pipeline used for data processing.
Defaults to None.
modality (dict, optional): Modality to specify the sensor data used
pipeline (List[dict]): Pipeline used for data processing.
Defaults to [].
modality (dict): Modality to specify the sensor data used
as input. Defaults to dict(use_lidar=True).
default_cam_key (str, optional): Default camera key for lidar2img
default_cam_key (str): Default camera key for lidar2img
association. Defaults to 'CAM_FRONT'.
box_type_3d (str, optional): Type of 3D box of this dataset.
box_type_3d (str): Type of 3D box of this dataset.
Based on the `box_type_3d`, the dataset will encapsulate the box
to its original format then converted them to `box_type_3d`.
Defaults to 'LiDAR' in this dataset. Available options includes:
......@@ -45,24 +45,30 @@ class WaymoDataset(KittiDataset):
- 'LiDAR': Box in LiDAR coordinates.
- 'Depth': Box in depth coordinates, usually for indoor dataset.
- 'Camera': Box in camera coordinates.
filter_empty_gt (bool, optional): Whether to filter empty GT.
Defaults to True.
test_mode (bool, optional): Whether the dataset is in test mode.
load_type (str): Type of loading mode. Defaults to 'frame_based'.
- 'frame_based': Load all of the instances in the frame.
- 'mv_image_based': Load all of the instances in the frame and need
to convert to the FOV-based data type to support image-based
detector.
- 'fov_image_based': Only load the instances inside the default
cam, and need to convert to the FOV-based data type to support
image-based detector.
filter_empty_gt (bool): Whether to filter the data with empty GT.
If it's set to be True, the example with empty annotations after
data pipeline will be dropped and a random example will be chosen
in `__getitem__`. Defaults to True.
test_mode (bool): Whether the dataset is in test mode.
Defaults to False.
pcd_limit_range (list[float], optional): The range of point cloud
pcd_limit_range (List[float]): The range of point cloud
used to filter invalid predicted boxes.
Defaults to [-85, -85, -5, 85, 85, 5].
cam_sync_instances (bool, optional): If use the camera sync label
cam_sync_instances (bool): If use the camera sync label
supported from waymo version 1.3.1. Defaults to False.
load_interval (int, optional): load frame interval.
Defaults to 1.
task (str, optional): task for 3D detection (lidar, mono3d).
lidar: take all the ground trurh in the frame.
mono3d: take the groundtruth that can be seen in the cam.
Defaults to 'lidar'.
max_sweeps (int, optional): max sweep for each frame. Defaults to 0.
load_interval (int): load frame interval. Defaults to 1.
max_sweeps (int): max sweep for each frame. Defaults to 0.
"""
METAINFO = {'CLASSES': ('Car', 'Pedestrian', 'Cyclist')}
METAINFO = {'classes': ('Car', 'Pedestrian', 'Cyclist')}
def __init__(self,
data_root: str,
......@@ -75,28 +81,27 @@ class WaymoDataset(KittiDataset):
CAM_SIDE_RIGHT='image_3',
CAM_SIDE_LEFT='image_4'),
pipeline: List[Union[dict, Callable]] = [],
modality: Optional[dict] = dict(use_lidar=True),
modality: dict = dict(use_lidar=True),
default_cam_key: str = 'CAM_FRONT',
box_type_3d: str = 'LiDAR',
load_type: str = 'frame_based',
filter_empty_gt: bool = True,
test_mode: bool = False,
pcd_limit_range: List[float] = [0, -40, -3, 70.4, 40, 0.0],
cam_sync_instances=False,
load_interval=1,
task='lidar_det',
max_sweeps=0,
**kwargs):
cam_sync_instances: bool = False,
load_interval: int = 1,
max_sweeps: int = 0,
**kwargs) -> None:
self.load_interval = load_interval
# set loading mode for different task settings
self.cam_sync_instances = cam_sync_instances
# construct self.cat_ids for vision-only anns parsing
self.cat_ids = range(len(self.METAINFO['CLASSES']))
self.cat_ids = range(len(self.METAINFO['classes']))
self.cat2label = {cat_id: i for i, cat_id in enumerate(self.cat_ids)}
self.max_sweeps = max_sweeps
self.task = task
# we do not provide file_client_args to custom_3d init
# because we want disk loading for info
# while ceph loading for KITTI2Waymo
# while ceph loading for Prediction2Waymo
super().__init__(
data_root=data_root,
ann_file=ann_file,
......@@ -108,16 +113,17 @@ class WaymoDataset(KittiDataset):
default_cam_key=default_cam_key,
data_prefix=data_prefix,
test_mode=test_mode,
load_type=load_type,
**kwargs)
def parse_ann_info(self, info: dict) -> dict:
"""Get annotation info according to the given index.
"""Process the `instances` in data info to `ann_info`.
Args:
info (dict): Data information of single data sample.
Returns:
dict: annotation information consists of the following keys:
dict: Annotation information consists of the following keys:
- bboxes_3d (:obj:`LiDARInstance3DBoxes`):
3D ground truth bboxes.
......@@ -150,7 +156,7 @@ class WaymoDataset(KittiDataset):
centers_2d = np.zeros((0, 2), dtype=np.float32)
depths = np.zeros((0), dtype=np.float32)
if self.task == 'mono_det':
if self.load_type in ['fov_image_based', 'mv_image_based']:
gt_bboxes_3d = CameraInstance3DBoxes(
ann_info['gt_bboxes_3d'],
box_dim=ann_info['gt_bboxes_3d'].shape[-1],
......@@ -182,13 +188,22 @@ class WaymoDataset(KittiDataset):
data_list = data_list[::self.load_interval]
return data_list
def parse_data_info(self, info: dict) -> dict:
def parse_data_info(self, info: dict) -> Union[dict, List[dict]]:
"""if task is lidar or multiview det, use super() method elif task is
mono3d, split the info from frame-wise to img-wise."""
if self.task != 'mono_det':
if self.cam_sync_instances:
# use the cam sync labels
info['instances'] = info['cam_sync_instances']
if self.load_type == 'frame_based':
return super().parse_data_info(info)
elif self.load_type == 'fov_image_based':
# only loading the fov image and the fov instance
new_image_info = {}
new_image_info[self.default_cam_key] = \
info['images'][self.default_cam_key]
info['images'] = new_image_info
info['instances'] = info['cam_instances'][self.default_cam_key]
return super().parse_data_info(info)
else:
# in the mono3d, the instances is from cam sync.
......
mmdet3d/evaluation/functional/waymo_utils/__init__.py
View file @ d7067e44
# Copyright (c) OpenMMLab. All rights reserved.
from .prediction_kitti_to_waymo import KITTI2Waymo
from .prediction_to_waymo import Prediction2Waymo
__all__ = ['KITTI2Waymo']
__all__ = ['Prediction2Waymo']
mmdet3d/evaluation/functional/waymo_utils/prediction_kitti_to_waymo.py → mmdet3d/evaluation/functional/waymo_utils/prediction_to_waymo.py
View file @ d7067e44
......@@ -5,29 +5,33 @@ r"""Adapted from `Waymo to KITTI converter
try:
from waymo_open_dataset import dataset_pb2 as open_dataset
from waymo_open_dataset import label_pb2
from waymo_open_dataset.protos import metrics_pb2
from waymo_open_dataset.protos.metrics_pb2 import Objects
except ImportError:
Objects = None
raise ImportError(
'Please run "pip install waymo-open-dataset-tf-2-1-0==1.2.0" '
'to install the official devkit first.')
from glob import glob
from os.path import join
from typing import List, Optional
import mmengine
import numpy as np
import tensorflow as tf
from waymo_open_dataset import label_pb2
from waymo_open_dataset.protos import metrics_pb2
class KITTI2Waymo(object):
"""KITTI predictions to Waymo converter.
class Prediction2Waymo(object):
"""Predictions to Waymo converter. The format of prediction results could
be original format or kitti-format.
This class serves as the converter to change predictions from KITTI to
Waymo format.
Args:
kitti_result_files (list[dict]): Predictions in KITTI format.
results (list[dict]): Prediction results.
waymo_tfrecords_dir (str): Directory to load waymo raw data.
waymo_results_save_dir (str): Directory to save converted predictions
in waymo format (.bin files).
......@@ -35,33 +39,47 @@ class KITTI2Waymo(object):
predictions in waymo format (.bin file), like 'a/b/c.bin'.
prefix (str): Prefix of filename. In general, 0 for training, 1 for
validation and 2 for testing.
workers (str): Number of parallel processes.
classes (dict): A list of class name.
workers (str): Number of parallel processes. Defaults to 2.
file_client_args (str): File client for reading gt in waymo format.
Defaults to ``dict(backend='disk')``.
from_kitti_format (bool, optional): Whether the reuslts are kitti
format. Defaults to False.
idx2metainfo (Optional[dict], optional): The mapping from sample_idx to
metainfo. The metainfo must contain the keys: 'idx2contextname' and
'idx2timestamp'. Defaults to None.
"""
def __init__(self,
kitti_result_files,
waymo_tfrecords_dir,
waymo_results_save_dir,
waymo_results_final_path,
prefix,
workers=64,
file_client_args=dict(backend='disk')):
self.kitti_result_files = kitti_result_files
results: List[dict],
waymo_tfrecords_dir: str,
waymo_results_save_dir: str,
waymo_results_final_path: str,
prefix: str,
classes: dict,
workers: int = 2,
file_client_args: dict = dict(backend='disk'),
from_kitti_format: bool = False,
idx2metainfo: Optional[dict] = None):
self.results = results
self.waymo_tfrecords_dir = waymo_tfrecords_dir
self.waymo_results_save_dir = waymo_results_save_dir
self.waymo_results_final_path = waymo_results_final_path
self.prefix = prefix
self.classes = classes
self.workers = int(workers)
self.file_client_args = file_client_args
self.name2idx = {}
for idx, result in enumerate(kitti_result_files):
if len(result['sample_id']) > 0:
self.name2idx[str(result['sample_id'][0])] = idx
self.from_kitti_format = from_kitti_format
if idx2metainfo is not None:
self.idx2metainfo = idx2metainfo
# If ``fast_eval``, the metainfo does not need to be read from
# original data online. It's preprocessed offline.
self.fast_eval = True
else:
self.fast_eval = False
# turn on eager execution for older tensorflow versions
if int(tf.__version__.split('.')[0]) < 2:
tf.enable_eager_execution()
self.name2idx = {}
self.k2w_cls_map = {
'Car': label_pb2.Label.TYPE_VEHICLE,
......@@ -70,12 +88,28 @@ class KITTI2Waymo(object):
'Cyclist': label_pb2.Label.TYPE_CYCLIST,
}
if self.from_kitti_format:
self.T_ref_to_front_cam = np.array([[0.0, 0.0, 1.0, 0.0],
[-1.0, 0.0, 0.0, 0.0],
[0.0, -1.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 1.0]])
# ``sample_idx`` of the sample in kitti-format is an array
for idx, result in enumerate(results):
if len(result['sample_idx']) > 0:
self.name2idx[str(result['sample_idx'][0])] = idx
else:
# ``sample_idx`` of the sample in the original prediction
# is an int value.
for idx, result in enumerate(results):
self.name2idx[str(result['sample_idx'])] = idx
if not self.fast_eval:
# need to read original '.tfrecord' file
self.get_file_names()
# turn on eager execution for older tensorflow versions
if int(tf.__version__.split('.')[0]) < 2:
tf.enable_eager_execution()
self.create_folder()
def get_file_names(self):
......@@ -192,6 +226,13 @@ class KITTI2Waymo(object):
file_idx (int): Index of the file to be converted.
"""
file_pathname = self.waymo_tfrecord_pathnames[file_idx]
if 's3://' in file_pathname and tf.__version__ >= '2.6.0':
try:
import tensorflow_io as tfio # noqa: F401
except ImportError:
raise ImportError(
"Please run 'pip install tensorflow-io' to install tensorflow_io first." # noqa: E501
)
file_data = tf.data.TFRecordDataset(file_pathname, compression_type='')
for frame_num, frame_data in enumerate(file_data):
......@@ -200,6 +241,11 @@ class KITTI2Waymo(object):
filename = f'{self.prefix}{file_idx:03d}{frame_num:03d}'
context_name = frame.context.name
frame_timestamp_micros = frame.timestamp_micros
if filename in self.name2idx:
if self.from_kitti_format:
for camera in frame.context.camera_calibrations:
# FRONT = 1, see dataset.proto for details
if camera.name == 1:
......@@ -208,14 +254,17 @@ class KITTI2Waymo(object):
T_k2w = T_front_cam_to_vehicle @ self.T_ref_to_front_cam
context_name = frame.context.name
frame_timestamp_micros = frame.timestamp_micros
if filename in self.name2idx:
kitti_result = \
self.kitti_result_files[self.name2idx[filename]]
objects = self.parse_objects(kitti_result, T_k2w, context_name,
self.results[self.name2idx[filename]]
objects = self.parse_objects(kitti_result, T_k2w,
context_name,
frame_timestamp_micros)
else:
index = self.name2idx[filename]
objects = self.parse_objects_from_origin(
self.results[index], context_name,
frame_timestamp_micros)
else:
print(filename, 'not found.')
objects = metrics_pb2.Objects()
......@@ -225,11 +274,100 @@ class KITTI2Waymo(object):
'wb') as f:
f.write(objects.SerializeToString())
def convert_one_fast(self, res_index: int):
"""Convert action for single file. It read the metainfo from the
preprocessed file offline and will be faster.
Args:
res_index (int): The indices of the results.
"""
sample_idx = self.results[res_index]['sample_idx']
if len(self.results[res_index]['pred_instances_3d']) > 0:
objects = self.parse_objects_from_origin(
self.results[res_index],
self.idx2metainfo[str(sample_idx)]['contextname'],
self.idx2metainfo[str(sample_idx)]['timestamp'])
else:
print(sample_idx, 'not found.')
objects = metrics_pb2.Objects()
with open(
join(self.waymo_results_save_dir, f'{sample_idx}.bin'),
'wb') as f:
f.write(objects.SerializeToString())
def parse_objects_from_origin(self, result: dict, contextname: str,
timestamp: str) -> Objects:
"""Parse obejcts from the original prediction results.
Args:
result (dict): The original prediction results.
contextname (str): The ``contextname`` of sample in waymo.
timestamp (str): The ``timestamp`` of sample in waymo.
Returns:
metrics_pb2.Objects: The parsed object.
"""
lidar_boxes = result['pred_instances_3d']['bboxes_3d'].tensor
scores = result['pred_instances_3d']['scores_3d']
labels = result['pred_instances_3d']['labels_3d']
def parse_one_object(index):
class_name = self.classes[labels[index].item()]
box = label_pb2.Label.Box()
height = lidar_boxes[index][5].item()
heading = lidar_boxes[index][6].item()
while heading < -np.pi:
heading += 2 * np.pi
while heading > np.pi:
heading -= 2 * np.pi
box.center_x = lidar_boxes[index][0].item()
box.center_y = lidar_boxes[index][1].item()
box.center_z = lidar_boxes[index][2].item() + height / 2
box.length = lidar_boxes[index][3].item()
box.width = lidar_boxes[index][4].item()
box.height = height
box.heading = heading
o = metrics_pb2.Object()
o.object.box.CopyFrom(box)
o.object.type = self.k2w_cls_map[class_name]
o.score = scores[index].item()
o.context_name = contextname
o.frame_timestamp_micros = timestamp
return o
objects = metrics_pb2.Objects()
for i in range(len(lidar_boxes)):
objects.objects.append(parse_one_object(i))
return objects
def convert(self):
"""Convert action."""
print('Start converting ...')
mmengine.track_parallel_progress(self.convert_one, range(len(self)),
self.workers)
convert_func = self.convert_one_fast if self.fast_eval else \
self.convert_one
# from torch.multiprocessing import set_sharing_strategy
# # Force using "file_system" sharing strategy for stability
# set_sharing_strategy("file_system")
# mmengine.track_parallel_progress(convert_func, range(len(self)),
# self.workers)
# TODO: Support multiprocessing. Now, multiprocessing evaluation will
# cause shared memory error in torch-1.10 and torch-1.11. Details can
# be seen in https://github.com/pytorch/pytorch/issues/67864.
prog_bar = mmengine.ProgressBar(len(self))
for i in range(len(self)):
convert_func(i)
prog_bar.update()
print('\nFinished ...')
# combine all files into one .bin
......@@ -241,7 +379,8 @@ class KITTI2Waymo(object):
def __len__(self):
"""Length of the filename list."""
return len(self.waymo_tfrecord_pathnames)
return len(self.results) if self.fast_eval else len(
self.waymo_tfrecord_pathnames)
def transform(self, T, x, y, z):
"""Transform the coordinates with matrix T.
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment