Skip to content

GitLab

"torchvision/git@developer.sourcefind.cn:OpenDAS/vision.git" did not exist on "2bababf2f285235cd113eea2462d09fa5aa845f6"
Commit 9fcfd78f authored by zhangwenwei's avatar zhangwenwei
Browse files

Merge branch 'regnet' into 'master' 

Regnet

See merge request open-mmlab/mmdet.3d!67
parents 0bd814d7 d5f4ece4
Hide whitespace changes
Inline Side-by-side
Showing with 897 additions and 8 deletions
configs/nus/hv_pointpillars_fpn_sbn-all_4x8_3x_nus-3d.py 0 → 100644
View file @ 9fcfd78f
# model settings
voxel_size = [0.25, 0.25, 8]
point_cloud_range = [-50, -50, -5, 50, 50, 3]
class_names = [
'car', 'truck', 'trailer', 'bus', 'construction_vehicle', 'bicycle',
'motorcycle', 'pedestrian', 'traffic_cone', 'barrier'
]
model = dict(
type='MVXFasterRCNNV2',
pts_voxel_layer=dict(
max_num_points=64,
point_cloud_range=point_cloud_range,
voxel_size=voxel_size,
max_voxels=(30000, 40000)),
pts_voxel_encoder=dict(
type='HardVFE',
in_channels=4,
feat_channels=[64, 64],
with_distance=False,
voxel_size=voxel_size,
with_cluster_center=True,
with_voxel_center=True,
point_cloud_range=point_cloud_range,
norm_cfg=dict(type='naiveSyncBN1d', eps=1e-3, momentum=0.01)),
pts_middle_encoder=dict(
type='PointPillarsScatter', in_channels=64, output_shape=[400, 400]),
pts_backbone=dict(
type='SECOND',
in_channels=64,
norm_cfg=dict(type='naiveSyncBN2d', eps=1e-3, momentum=0.01),
layer_nums=[3, 5, 5],
layer_strides=[2, 2, 2],
out_channels=[64, 128, 256]),
pts_neck=dict(
type='FPN',
norm_cfg=dict(type='naiveSyncBN2d', eps=1e-3, momentum=0.01),
act_cfg=dict(type='ReLU'),
in_channels=[64, 128, 256],
out_channels=256,
start_level=0,
num_outs=3),
pts_bbox_head=dict(
type='Anchor3DHead',
num_classes=10,
in_channels=256,
feat_channels=256,
use_direction_classifier=True,
anchor_generator=dict(
type='AlignedAnchor3DRangeGenerator',
ranges=[[-50, -50, -1.8, 50, 50, -1.8]],
scales=[1, 2, 4],
sizes=[
[0.8660, 2.5981, 1.], # 1.5/sqrt(3)
[0.5774, 1.7321, 1.], # 1/sqrt(3)
[1., 1., 1.],
[0.4, 0.4, 1],
],
custom_values=[0, 0],
rotations=[0, 1.57],
reshape_out=True),
assigner_per_size=False,
diff_rad_by_sin=True,
dir_offset=0.7854, # pi/4
dir_limit_offset=0,
bbox_coder=dict(type='DeltaXYZWLHRBBoxCoder', code_size=9),
loss_cls=dict(
type='FocalLoss',
use_sigmoid=True,
gamma=2.0,
alpha=0.25,
loss_weight=1.0),
loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0),
loss_dir=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.2)))
# model training and testing settings
train_cfg = dict(
pts=dict(
assigner=dict( # for Car
type='MaxIoUAssigner',
iou_calculator=dict(type='BboxOverlapsNearest3D'),
pos_iou_thr=0.6,
neg_iou_thr=0.3,
min_pos_iou=0.3,
ignore_iof_thr=-1),
allowed_border=0,
code_weight=[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.2, 0.2],
pos_weight=-1,
debug=False))
test_cfg = dict(
pts=dict(
use_rotate_nms=True,
nms_across_levels=False,
nms_pre=1000,
nms_thr=0.2,
score_thr=0.05,
min_bbox_size=0,
max_num=500))
# dataset settings
dataset_type = 'NuScenesDataset'
data_root = 'data/nuscenes/'
img_norm_cfg = dict(
mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)
input_modality = dict(
use_lidar=True,
use_depth=False,
use_lidar_intensity=True,
use_camera=False)
file_client_args = dict(backend='disk')
# file_client_args = dict(
# backend='petrel',
# path_mapping=dict({
# './data/nuscenes/': 's3://nuscenes/nuscenes/',
# 'data/nuscenes/': 's3://nuscenes/nuscenes/'
# }))
train_pipeline = [
dict(
type='LoadPointsFromFile',
load_dim=5,
use_dim=5,
file_client_args=file_client_args),
dict(
type='LoadPointsFromMultiSweeps',
sweeps_num=10,
file_client_args=file_client_args),
dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True),
dict(
type='GlobalRotScale',
rot_uniform_noise=[-0.3925, 0.3925],
scaling_uniform_noise=[0.95, 1.05],
trans_normal_noise=[0, 0, 0]),
dict(type='RandomFlip3D', flip_ratio=0.5),
dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
dict(type='PointShuffle'),
dict(type='DefaultFormatBundle3D', class_names=class_names),
dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d'])
]
test_pipeline = [
dict(
type='LoadPointsFromFile',
load_dim=5,
use_dim=5,
file_client_args=file_client_args),
dict(
type='LoadPointsFromMultiSweeps',
sweeps_num=10,
file_client_args=file_client_args),
dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
dict(type='RandomFlip3D', flip_ratio=0),
dict(
type='DefaultFormatBundle3D',
class_names=class_names,
with_label=False),
dict(type='Collect3D', keys=['points'])
]
data = dict(
samples_per_gpu=4,
workers_per_gpu=4,
train=dict(
type=dataset_type,
data_root=data_root,
ann_file=data_root + 'nuscenes_infos_train.pkl',
pipeline=train_pipeline,
modality=input_modality,
classes=class_names,
test_mode=False),
val=dict(
type=dataset_type,
data_root=data_root,
ann_file=data_root + 'nuscenes_infos_val.pkl',
pipeline=test_pipeline,
modality=input_modality,
classes=class_names,
test_mode=True),
test=dict(
type=dataset_type,
data_root=data_root,
ann_file=data_root + 'nuscenes_infos_val.pkl',
pipeline=test_pipeline,
modality=input_modality,
classes=class_names,
test_mode=True))
# optimizer
optimizer = dict(type='AdamW', lr=0.001, weight_decay=0.01)
# max_norm=10 is better for SECOND
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=1000,
warmup_ratio=1.0 / 1000,
step=[28, 34])
momentum_config = None
checkpoint_config = dict(interval=1)
# yapf:disable
evaluation = dict(interval=36)
log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook'),
dict(type='TensorboardLoggerHook')
])
# yapf:enable
# runtime settings
total_epochs = 36
dist_params = dict(backend='nccl')
log_level = 'INFO'
work_dir = './work_dirs/hv_pointpillars_fpn_sbn-all_4x8_2x_nus-3d'
load_from = None
resume_from = None
workflow = [('train', 1)]
configs/regnet/README.md 0 → 100644
View file @ 9fcfd78f
# Designing Network Design Spaces
## Introduction
We implement RegNetX and RegNetY models in 3D detection systems and provide their first results on PointPillars.
The pre-trained modles are converted from [model zoo of pycls](https://github.com/facebookresearch/pycls/blob/master/MODEL_ZOO.md) and maintained in [mmcv](https://github.com/open-mmlab/mmcv).
```
@article{radosavovic2020designing,
title={Designing Network Design Spaces},
author={Ilija Radosavovic and Raj Prateek Kosaraju and Ross Girshick and Kaiming He and Piotr Dollár},
year={2020},
eprint={2003.13678},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
## Usage
To use a regnet model, there are two steps to do:
1. Convert the model to ResNet-style supported by MMDetection
2. Modify backbone and neck in config accordingly
### Convert model
We already prepare models of FLOPs from 800M to 12G in our model zoo.
For more general usage, we also provide script `regnet2mmdet.py` in the tools directory to convert the key of models pretrained by [pycls](https://github.com/facebookresearch/pycls/) to
ResNet-style checkpoints used in MMDetection.
```bash
python -u tools/regnet2mmdet.py ${PRETRAIN_PATH} ${STORE_PATH}
```
This script convert model from `PRETRAIN_PATH` and store the converted model in `STORE_PATH`.
### Modify config
The users can modify the config's `depth` of backbone and corresponding keys in `arch` according to the configs in the [pycls model zoo](https://github.com/facebookresearch/pycls/blob/master/MODEL_ZOO.md).
The parameter `in_channels` in FPN can be found in the Figure 15 & 16 of the paper (`wi` in the legend).
This directory already provides some configs with their performance, using RegNetX from 800MF to 12GF level.
For other pre-trained models or self-implemented regnet models, the users are responsible to check these parameters by themselves.
**Note**: Although Fig. 15 & 16 also provide `w0`, `wa`, `wm`, `group_w`, and `bot_mul` for `arch`, they are quantized thus inaccurate, using them sometimes produces different backbone that does not match the key in the pre-trained model.
## Results
### PointPillars
| Backbone | Lr schd | Mem (GB) | Inf time (fps) | mAP |NDS| Download |
| :---------: | :-----: | :------: | :------------: | :----: |:----: | :------: |
| [SECFPN](../) | 2x ||||
|[RegNetX-400MF-SECFPN](./hv_pointpillars_regnet-400mf_secfpn_sbn-all_4x8_2x_nus-3d.py)| 2x ||||
| [FPN](../) | 2x ||||
|[RegNetX-400MF-FPN](./hv_pointpillars_regnet-400mf_fpn_sbn-all_4x8_2x_nus-3d.py)| 2x ||||
configs/regnet/hv_pointpillars_regnet-400mf_fpn_sbn-all_4x8_2x_nus-3d.py 0 → 100644
View file @ 9fcfd78f
# model settings
voxel_size = [0.25, 0.25, 8]
point_cloud_range = [-50, -50, -5, 50, 50, 3]
class_names = [
'car', 'truck', 'trailer', 'bus', 'construction_vehicle', 'bicycle',
'motorcycle', 'pedestrian', 'traffic_cone', 'barrier'
]
model = dict(
type='MVXFasterRCNNV2',
pretrained=dict(pts='open-mmlab://regnetx_400mf'),
pts_voxel_layer=dict(
max_num_points=64,
point_cloud_range=point_cloud_range,
voxel_size=voxel_size,
max_voxels=(30000, 40000)),
pts_voxel_encoder=dict(
type='HardVFE',
in_channels=4,
feat_channels=[64, 64],
with_distance=False,
voxel_size=voxel_size,
with_cluster_center=True,
with_voxel_center=True,
point_cloud_range=point_cloud_range,
norm_cfg=dict(type='naiveSyncBN1d', eps=1e-3, momentum=0.01)),
pts_middle_encoder=dict(
type='PointPillarsScatter', in_channels=64, output_shape=[400, 400]),
pts_backbone=dict(
type='NoStemRegNet',
arch=dict(w0=24, wa=24.48, wm=2.54, group_w=16, depth=22, bot_mul=1.0),
out_indices=(1, 2, 3),
frozen_stages=-1,
strides=(1, 2, 2, 2),
base_channels=64,
norm_cfg=dict(type='naiveSyncBN2d', eps=1e-3, momentum=0.01),
norm_eval=False,
style='pytorch'),
pts_neck=dict(
type='FPN',
norm_cfg=dict(type='naiveSyncBN2d', eps=1e-3, momentum=0.01),
act_cfg=dict(type='ReLU'),
in_channels=[64, 160, 384],
out_channels=256,
start_level=0,
num_outs=3),
pts_bbox_head=dict(
type='Anchor3DHead',
num_classes=10,
in_channels=256,
feat_channels=256,
use_direction_classifier=True,
anchor_generator=dict(
type='AlignedAnchor3DRangeGenerator',
ranges=[[-50, -50, -1.8, 50, 50, -1.8]],
scales=[1, 2, 4],
sizes=[
[0.8660, 2.5981, 1.], # 1.5/sqrt(3)
[0.5774, 1.7321, 1.], # 1/sqrt(3)
[1., 1., 1.],
[0.4, 0.4, 1],
],
custom_values=[0, 0],
rotations=[0, 1.57],
reshape_out=True),
assigner_per_size=False,
diff_rad_by_sin=True,
dir_offset=0.7854, # pi/4
dir_limit_offset=0,
bbox_coder=dict(type='DeltaXYZWLHRBBoxCoder', code_size=9),
loss_cls=dict(
type='FocalLoss',
use_sigmoid=True,
gamma=2.0,
alpha=0.25,
loss_weight=1.0),
loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0),
loss_dir=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.2)))
# model training and testing settings
train_cfg = dict(
pts=dict(
assigner=dict(
type='MaxIoUAssigner',
iou_calculator=dict(type='BboxOverlapsNearest3D'),
pos_iou_thr=0.6,
neg_iou_thr=0.3,
min_pos_iou=0.3,
ignore_iof_thr=-1),
allowed_border=0,
code_weight=[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.2, 0.2],
pos_weight=-1,
debug=False))
test_cfg = dict(
pts=dict(
use_rotate_nms=True,
nms_across_levels=False,
nms_pre=1000,
nms_thr=0.2,
score_thr=0.05,
min_bbox_size=0,
max_num=500))
# dataset settings
dataset_type = 'NuScenesDataset'
data_root = 'data/nuscenes/'
input_modality = dict(
use_lidar=True,
use_depth=False,
use_lidar_intensity=True,
use_camera=False)
file_client_args = dict(backend='disk')
# file_client_args = dict(
# backend='petrel',
# path_mapping=dict({
# './data/nuscenes/': 's3://nuscenes/nuscenes/',
# 'data/nuscenes/': 's3://nuscenes/nuscenes/'
# }))
train_pipeline = [
dict(
type='LoadPointsFromFile',
load_dim=5,
use_dim=5,
file_client_args=file_client_args),
dict(
type='LoadPointsFromMultiSweeps',
sweeps_num=10,
file_client_args=file_client_args),
dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True),
dict(
type='GlobalRotScale',
rot_uniform_noise=[-0.3925, 0.3925],
scaling_uniform_noise=[0.95, 1.05],
trans_normal_noise=[0, 0, 0]),
dict(type='RandomFlip3D', flip_ratio=0.5),
dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
dict(type='PointShuffle'),
dict(type='DefaultFormatBundle3D', class_names=class_names),
dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d'])
]
test_pipeline = [
dict(
type='LoadPointsFromFile',
load_dim=5,
use_dim=5,
file_client_args=file_client_args),
dict(
type='LoadPointsFromMultiSweeps',
sweeps_num=10,
file_client_args=file_client_args),
dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
dict(type='RandomFlip3D', flip_ratio=0),
dict(
type='DefaultFormatBundle3D',
class_names=class_names,
with_label=False),
dict(type='Collect3D', keys=['points'])
]
data = dict(
samples_per_gpu=4,
workers_per_gpu=4,
train=dict(
type=dataset_type,
data_root=data_root,
ann_file=data_root + 'nuscenes_infos_train.pkl',
pipeline=train_pipeline,
modality=input_modality,
classes=class_names,
test_mode=False),
val=dict(
type=dataset_type,
data_root=data_root,
ann_file=data_root + 'nuscenes_infos_val.pkl',
pipeline=test_pipeline,
modality=input_modality,
classes=class_names,
test_mode=True),
test=dict(
type=dataset_type,
data_root=data_root,
ann_file=data_root + 'nuscenes_infos_val.pkl',
pipeline=test_pipeline,
modality=input_modality,
classes=class_names,
test_mode=True))
# optimizer
optimizer = dict(type='AdamW', lr=0.001, weight_decay=0.01)
# max_norm=10 is better for SECOND
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=1000,
warmup_ratio=1.0 / 1000,
step=[20, 23])
momentum_config = None
checkpoint_config = dict(interval=1)
# yapf:disable
evaluation = dict(interval=24)
log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook'),
dict(type='TensorboardLoggerHook')
])
# yapf:enable
# runtime settings
total_epochs = 24
dist_params = dict(backend='nccl')
log_level = 'INFO'
work_dir = './work_dirs/hv_pointpillars_secfpn_sbn-all_4x8_2x_nus-3d'
load_from = None
resume_from = None
workflow = [('train', 1)]
configs/regnet/hv_pointpillars_regnet-400mf_secfpn_sbn-all_4x8_2x_nus-3d.py 0 → 100644
View file @ 9fcfd78f
# model settings
voxel_size = [0.25, 0.25, 8]
point_cloud_range = [-50, -50, -5, 50, 50, 3]
class_names = [
'car', 'truck', 'trailer', 'bus', 'construction_vehicle', 'bicycle',
'motorcycle', 'pedestrian', 'traffic_cone', 'barrier'
]
model = dict(
type='MVXFasterRCNNV2',
pretrained=dict(pts='open-mmlab://regnetx_400mf'),
pts_voxel_layer=dict(
max_num_points=64,
point_cloud_range=point_cloud_range,
voxel_size=voxel_size,
max_voxels=(30000, 40000)),
pts_voxel_encoder=dict(
type='HardVFE',
in_channels=4,
feat_channels=[64, 64],
with_distance=False,
voxel_size=voxel_size,
with_cluster_center=True,
with_voxel_center=True,
point_cloud_range=point_cloud_range,
norm_cfg=dict(type='naiveSyncBN1d', eps=1e-3, momentum=0.01)),
pts_middle_encoder=dict(
type='PointPillarsScatter', in_channels=64, output_shape=[400, 400]),
pts_backbone=dict(
type='NoStemRegNet',
arch=dict(w0=24, wa=24.48, wm=2.54, group_w=16, depth=22, bot_mul=1.0),
out_indices=(1, 2, 3),
frozen_stages=-1,
strides=(1, 2, 2, 2),
base_channels=64,
norm_cfg=dict(type='naiveSyncBN2d', eps=1e-3, momentum=0.01),
norm_eval=False,
style='pytorch'),
pts_neck=dict(
type='SECONDFPN',
norm_cfg=dict(type='naiveSyncBN2d', eps=1e-3, momentum=0.01),
in_channels=[64, 160, 384],
upsample_strides=[1, 2, 4],
out_channels=[128, 128, 128]),
pts_bbox_head=dict(
type='Anchor3DHead',
num_classes=10,
in_channels=384,
feat_channels=384,
use_direction_classifier=True,
anchor_generator=dict(
type='Anchor3DRangeGenerator',
ranges=[
[-49.6, -49.6, -1.80032795, 49.6, 49.6, -1.80032795],
[-49.6, -49.6, -1.74440365, 49.6, 49.6, -1.74440365],
[-49.6, -49.6, -1.68526504, 49.6, 49.6, -1.68526504],
[-49.6, -49.6, -1.67339111, 49.6, 49.6, -1.67339111],
[-49.6, -49.6, -1.61785072, 49.6, 49.6, -1.61785072],
[-49.6, -49.6, -1.80984986, 49.6, 49.6, -1.80984986],
[-49.6, -49.6, -1.763965, 49.6, 49.6, -1.763965],
],
sizes=[
[1.95017717, 4.60718145, 1.72270761], # car
[2.4560939, 6.73778078, 2.73004906], # truck
[2.87427237, 12.01320693, 3.81509561], # trailer
[0.60058911, 1.68452161, 1.27192197], # bicycle
[0.66344886, 0.7256437, 1.75748069], # pedestrian
[0.39694519, 0.40359262, 1.06232151], # traffic_cone
[2.49008838, 0.48578221, 0.98297065], # barrier
],
custom_values=[0, 0],
rotations=[0, 1.57],
reshape_out=True),
assigner_per_size=False,
diff_rad_by_sin=True,
dir_offset=0.7854, # pi/4
dir_limit_offset=0,
bbox_coder=dict(type='DeltaXYZWLHRBBoxCoder', code_size=9),
loss_cls=dict(
type='FocalLoss',
use_sigmoid=True,
gamma=2.0,
alpha=0.25,
loss_weight=1.0),
loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0),
loss_dir=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.2)))
# model training and testing settings
train_cfg = dict(
pts=dict(
assigner=dict( # for Car
type='MaxIoUAssigner',
iou_calculator=dict(type='BboxOverlapsNearest3D'),
pos_iou_thr=0.6,
neg_iou_thr=0.3,
min_pos_iou=0.3,
ignore_iof_thr=-1),
allowed_border=0,
code_weight=[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.2, 0.2],
pos_weight=-1,
debug=False))
test_cfg = dict(
pts=dict(
use_rotate_nms=True,
nms_across_levels=False,
nms_pre=1000,
nms_thr=0.2,
score_thr=0.05,
min_bbox_size=0,
max_num=500))
# dataset settings
dataset_type = 'NuScenesDataset'
data_root = 'data/nuscenes/'
input_modality = dict(
use_lidar=True,
use_depth=False,
use_lidar_intensity=True,
use_camera=False,
)
file_client_args = dict(backend='disk')
# file_client_args = dict(
# backend='petrel',
# path_mapping=dict({
# './data/nuscenes/': 's3://nuscenes/nuscenes/',
# 'data/nuscenes/': 's3://nuscenes/nuscenes/'
# }))
train_pipeline = [
dict(
type='LoadPointsFromFile',
load_dim=5,
use_dim=5,
file_client_args=file_client_args),
dict(
type='LoadPointsFromMultiSweeps',
sweeps_num=10,
file_client_args=file_client_args),
dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True),
dict(
type='GlobalRotScale',
rot_uniform_noise=[-0.3925, 0.3925],
scaling_uniform_noise=[0.95, 1.05],
trans_normal_noise=[0, 0, 0]),
dict(type='RandomFlip3D', flip_ratio=0.5),
dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
dict(type='PointShuffle'),
dict(type='DefaultFormatBundle3D', class_names=class_names),
dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d']),
]
test_pipeline = [
dict(
type='LoadPointsFromFile',
load_dim=5,
use_dim=5,
file_client_args=file_client_args),
dict(
type='LoadPointsFromMultiSweeps',
sweeps_num=10,
file_client_args=file_client_args),
dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
dict(type='RandomFlip3D', flip_ratio=0),
dict(
type='DefaultFormatBundle3D',
class_names=class_names,
with_label=False),
dict(type='Collect3D', keys=['points']),
]
data = dict(
samples_per_gpu=4,
workers_per_gpu=4,
train=dict(
type=dataset_type,
data_root=data_root,
ann_file=data_root + 'nuscenes_infos_train.pkl',
pipeline=train_pipeline,
modality=input_modality,
classes=class_names,
test_mode=False),
val=dict(
type=dataset_type,
data_root=data_root,
ann_file=data_root + 'nuscenes_infos_val.pkl',
pipeline=test_pipeline,
modality=input_modality,
classes=class_names,
test_mode=True),
test=dict(
type=dataset_type,
data_root=data_root,
ann_file=data_root + 'nuscenes_infos_val.pkl',
pipeline=test_pipeline,
modality=input_modality,
classes=class_names,
test_mode=True))
# optimizer
optimizer = dict(type='AdamW', lr=0.001, weight_decay=0.01)
# max_norm=10 is better for SECOND
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=1000,
warmup_ratio=1.0 / 1000,
step=[20, 23])
momentum_config = None
checkpoint_config = dict(interval=1)
# yapf:disable
evaluation = dict(interval=24)
log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook'),
dict(type='TensorboardLoggerHook')
])
# yapf:enable
# runtime settings
total_epochs = 24
dist_params = dict(backend='nccl')
log_level = 'INFO'
work_dir = './work_dirs/hv_pointpillars_secfpn_sbn-all_4x8_2x_nus-3d'
load_from = None
resume_from = None
workflow = [('train', 1)]
mmdet3d/models/backbones/__init__.py
View file @ 9fcfd78f
from mmdet.models.backbones import SSDVGG, HRNet, ResNet, ResNetV1d, ResNeXt from mmdet.models.backbones import SSDVGG, HRNet, ResNet, ResNetV1d, ResNeXt
from .nostem_regnet import NoStemRegNet
from .pointnet2_sa_ssg import PointNet2SASSG from .pointnet2_sa_ssg import PointNet2SASSG
from .second import SECOND from .second import SECOND
__all__ = [ __all__ = [
'ResNet', 'ResNetV1d', 'ResNeXt', 'SSDVGG', 'HRNet', 'SECOND', 'ResNet', 'ResNetV1d', 'ResNeXt', 'SSDVGG', 'HRNet', 'NoStemRegNet',
'PointNet2SASSG' 'SECOND', 'PointNet2SASSG'
] ]
mmdet3d/models/backbones/nostem_regnet.py 0 → 100644
View file @ 9fcfd78f
from mmdet.models.backbones import RegNet
from ..builder import BACKBONES
@BACKBONES.register_module()
class NoStemRegNet(RegNet):
"""RegNet backbone without Stem for 3D detection.
More details can be found in `paper <https://arxiv.org/abs/2003.13678>`_ .
Args:
arch (dict): The parameter of RegNets.
- w0 (int): initial width
- wa (float): slope of width
- wm (float): quantization parameter to quantize the width
- depth (int): depth of the backbone
- group_w (int): width of group
- bot_mul (float): bottleneck ratio, i.e. expansion of bottlneck.
strides (Sequence[int]): Strides of the first block of each stage.
base_channels (int): Base channels after stem layer.
in_channels (int): Number of input image channels. Normally 3.
dilations (Sequence[int]): Dilation of each stage.
out_indices (Sequence[int]): Output from which stages.
style (str): `pytorch` or `caffe`. If set to "pytorch", the stride-two
layer is the 3x3 conv layer, otherwise the stride-two layer is
the first 1x1 conv layer.
frozen_stages (int): Stages to be frozen (all param fixed). -1 means
not freezing any parameters.
norm_cfg (dict): dictionary to construct and config norm layer.
norm_eval (bool): Whether to set norm layers to eval mode, namely,
freeze running stats (mean and var). Note: Effect on Batch Norm
and its variants only.
with_cp (bool): Use checkpoint or not. Using checkpoint will save some
memory while slowing down the training speed.
zero_init_residual (bool): whether to use zero init for last norm layer
in resblocks to let them behave as identity.
Example:
>>> from mmdet3d.models import NoStemRegNet
>>> import torch
>>> self = NoStemRegNet(
arch=dict(
w0=88,
wa=26.31,
wm=2.25,
group_w=48,
depth=25,
bot_mul=1.0))
>>> self.eval()
>>> inputs = torch.rand(1, 64, 16, 16)
>>> level_outputs = self.forward(inputs)
>>> for level_out in level_outputs:
... print(tuple(level_out.shape))
(1, 96, 8, 8)
(1, 192, 4, 4)
(1, 432, 2, 2)
(1, 1008, 1, 1)
"""
def __init__(self, arch, **kwargs):
super(NoStemRegNet, self).__init__(arch, **kwargs)
def _make_stem_layer(self, in_channels, base_channels):
return
def forward(self, x):
outs = []
for i, layer_name in enumerate(self.res_layers):
res_layer = getattr(self, layer_name)
x = res_layer(x)
if i in self.out_indices:
outs.append(x)
return tuple(outs)
mmdet3d/models/detectors/mvx_two_stage.py
View file @ 9fcfd78f
...@@ -66,21 +66,30 @@ class MVXTwoStageDetector(BaseDetector): ...@@ -66,21 +66,30 @@ class MVXTwoStageDetector(BaseDetector):
def init_weights(self, pretrained=None): def init_weights(self, pretrained=None):
super(MVXTwoStageDetector, self).init_weights(pretrained) super(MVXTwoStageDetector, self).init_weights(pretrained)
if pretrained is None:
img_pretrained = None
pts_pretrained = None
elif isinstance(pretrained, dict):
img_pretrained = pretrained.get('img', None)
pts_pretrained = pretrained.get('pts', None)
else:
raise ValueError(
f'pretrained should be a dict, got {type(pretrained)}')
if self.with_img_backbone: if self.with_img_backbone:
self.img_backbone.init_weights(pretrained=pretrained) self.img_backbone.init_weights(pretrained=img_pretrained)
if self.with_pts_backbone:
self.pts_backbone.init_weights(pretrained=pts_pretrained)
if self.with_img_neck: if self.with_img_neck:
if isinstance(self.img_neck, nn.Sequential): if isinstance(self.img_neck, nn.Sequential):
for m in self.img_neck: for m in self.img_neck:
m.init_weights() m.init_weights()
else: else:
self.img_neck.init_weights() self.img_neck.init_weights()
if self.with_shared_head:
self.img_shared_head.init_weights(pretrained=pretrained) if self.with_img_roi_head:
self.img_roi_head.init_weights(img_pretrained)
if self.with_img_rpn: if self.with_img_rpn:
self.img_rpn_head.init_weights() self.img_rpn_head.init_weights()
if self.with_img_bbox:
self.img_bbox_roi_extractor.init_weights()
self.img_bbox_head.init_weights()
if self.with_pts_bbox: if self.with_pts_bbox:
self.pts_bbox_head.init_weights() self.pts_bbox_head.init_weights()
...@@ -103,6 +112,10 @@ class MVXTwoStageDetector(BaseDetector): ...@@ -103,6 +112,10 @@ class MVXTwoStageDetector(BaseDetector):
def with_img_backbone(self): def with_img_backbone(self):
return hasattr(self, 'img_backbone') and self.img_backbone is not None return hasattr(self, 'img_backbone') and self.img_backbone is not None
@property
def with_pts_backbone(self):
return hasattr(self, 'pts_backbone') and self.pts_backbone is not None
@property @property
def with_fusion(self): def with_fusion(self):
return hasattr(self, return hasattr(self,
...@@ -120,6 +133,10 @@ class MVXTwoStageDetector(BaseDetector): ...@@ -120,6 +133,10 @@ class MVXTwoStageDetector(BaseDetector):
def with_img_rpn(self): def with_img_rpn(self):
return hasattr(self, 'img_rpn_head') and self.img_rpn_head is not None return hasattr(self, 'img_rpn_head') and self.img_rpn_head is not None
@property
def with_img_roi_head(self):
return hasattr(self, 'img_roi_head') and self.img_roi_head is not None
def extract_img_feat(self, img, img_meta): def extract_img_feat(self, img, img_meta):
if self.with_img_backbone: if self.with_img_backbone:
if img.dim() == 5 and img.size(0) == 1: if img.dim() == 5 and img.size(0) == 1:
......
tools/regnet2mmdet.py 0 → 100644
View file @ 9fcfd78f
import argparse
from collections import OrderedDict
import torch
def convert_stem(model_key, model_weight, state_dict, converted_names):
new_key = model_key.replace('stem.conv', 'conv1')
new_key = new_key.replace('stem.bn', 'bn1')
state_dict[new_key] = model_weight
converted_names.add(model_key)
print(f'Convert {model_key} to {new_key}')
def convert_head(model_key, model_weight, state_dict, converted_names):
new_key = model_key.replace('head.fc', 'fc')
state_dict[new_key] = model_weight
converted_names.add(model_key)
print(f'Convert {model_key} to {new_key}')
def convert_reslayer(model_key, model_weight, state_dict, converted_names):
split_keys = model_key.split('.')
layer, block, module = split_keys[:3]
block_id = int(block[1:])
layer_name = f'layer{int(layer[1:])}'
block_name = f'{block_id - 1}'
if block_id == 1 and module == 'bn':
new_key = f'{layer_name}.{block_name}.downsample.1.{split_keys[-1]}'
elif block_id == 1 and module == 'proj':
new_key = f'{layer_name}.{block_name}.downsample.0.{split_keys[-1]}'
elif module == 'f':
if split_keys[3] == 'a_bn':
module_name = 'bn1'
elif split_keys[3] == 'b_bn':
module_name = 'bn2'
elif split_keys[3] == 'c_bn':
module_name = 'bn3'
elif split_keys[3] == 'a':
module_name = 'conv1'
elif split_keys[3] == 'b':
module_name = 'conv2'
elif split_keys[3] == 'c':
module_name = 'conv3'
new_key = f'{layer_name}.{block_name}.{module_name}.{split_keys[-1]}'
else:
raise ValueError(f'Unsupported conversion of key {model_key}')
print(f'Convert {model_key} to {new_key}')
state_dict[new_key] = model_weight
converted_names.add(model_key)
def convert(src, dst):
"""Convert keys in pycls pretrained RegNet models to mmdet style."""
# load caffe model
regnet_model = torch.load(src)
blobs = regnet_model['model_state']
# convert to pytorch style
state_dict = OrderedDict()
converted_names = set()
for key, weight in blobs.items():
if 'stem' in key:
convert_stem(key, weight, state_dict, converted_names)
elif 'head' in key:
convert_head(key, weight, state_dict, converted_names)
elif key.startswith('s'):
convert_reslayer(key, weight, state_dict, converted_names)
# check if all layers are converted
for key in blobs:
if key not in converted_names:
print(f'not converted: {key}')
# save checkpoint
checkpoint = dict()
checkpoint['state_dict'] = state_dict
torch.save(checkpoint, dst)
def main():
parser = argparse.ArgumentParser(description='Convert model keys')
parser.add_argument('src', help='src detectron model path')
parser.add_argument('dst', help='save path')
args = parser.parse_args()
convert(args.src, args.dst)
if __name__ == '__main__':
main()
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