Merge pull request #683 from sshaoshuai/master

Release OpenPCDet v0.5.0

requirements.txt

-numpy
-torch>=1.1
+numpy<=1.20
+llvmlite
 numba
+torch>=1.1
 tensorboardX
 easydict
 pyyaml
 scikit-image
 tqdm
-kornia
 torchvision
+SharedArray

setup.py

 import os
-import sys
 import subprocess
 
 from setuptools import find_packages, setup
-from setuptools.command.install import install
-# TODO: This is a bit buggy since it requires torch before installing torch.
 from torch.utils.cpp_extension import BuildExtension, CUDAExtension
 
 
@@ -30,17 +27,8 @@ def write_version_to_file(version, target_file):
         print('__version__ = "%s"' % version, file=f)
 
 
-class PostInstallation(install):
-    """Post-installation for installation mode."""
-    def run(self):
-        install.run(self)
-        # Note: buggy for kornia==0.5.3 and it will be fixed in the next version.
-        # Set kornia to 0.5.2 temporarily
-        subprocess.call([sys.executable, '-m', 'pip', 'install', 'kornia==0.5.2', '--no-dependencies'])
-
-
 if __name__ == '__main__':
-    version = '0.4.0+%s' % get_git_commit_number()
+    version = '0.5.0+%s' % get_git_commit_number()
     write_version_to_file(version, 'pcdet/version.py')
 
     setup(
@@ -48,23 +36,24 @@ if __name__ == '__main__':
         version=version,
         description='OpenPCDet is a general codebase for 3D object detection from point cloud',
         install_requires=[
-            'numpy',
-            'torch>=1.1',
-            # 'spconv',  # spconv has different names depending on the cuda version
+            'numpy<=1.20',
+            'llvmlite',
             'numba',
             'tensorboardX',
             'easydict',
-            'pyyaml'
+            'pyyaml',
+            'scikit-image',
+            'tqdm',
+            'SharedArray',
+            # 'spconv',  # spconv has different names depending on the cuda version
         ],
+
         author='Shaoshuai Shi',
         author_email='shaoshuaics@gmail.com',
         license='Apache License 2.0',
         packages=find_packages(exclude=['tools', 'data', 'output']),
         cmdclass={
             'build_ext': BuildExtension,
-            'install': PostInstallation,
-            # Post installation cannot be done. ref: https://github.com/pypa/setuptools/issues/1936.
-            # 'develop': PostInstallation,
         },
         ext_modules=[
             make_cuda_ext(

tools/_init_path.py

+import sys
+sys.path.insert(0, '../')
\ No newline at end of file

tools/cfgs/dataset_configs/waymo_dataset.yaml

 DATASET: 'WaymoDataset'
 DATA_PATH: '../data/waymo'
-PROCESSED_DATA_TAG: 'waymo_processed_data'
+PROCESSED_DATA_TAG: 'waymo_processed_data_v0_5_0'
 
 POINT_CLOUD_RANGE: [-75.2, -75.2, -2, 75.2, 75.2, 4]
 
@@ -11,16 +11,27 @@ DATA_SPLIT: {
 
 SAMPLED_INTERVAL: {
     'train': 5,
-    'test': 5
+    'test': 1
 }
 
+FILTER_EMPTY_BOXES_FOR_TRAIN: True
+DISABLE_NLZ_FLAG_ON_POINTS: True
+
+USE_SHARED_MEMORY: False  # it will load the data to shared memory to speed up (DO NOT USE IT IF YOU DO NOT FULLY UNDERSTAND WHAT WILL HAPPEN)
+SHARED_MEMORY_FILE_LIMIT: 35000  # set it based on the size of your shared memory
+
 DATA_AUGMENTOR:
     DISABLE_AUG_LIST: ['placeholder']
     AUG_CONFIG_LIST:
         - NAME: gt_sampling
           USE_ROAD_PLANE: False
           DB_INFO_PATH:
-              - pcdet_waymo_dbinfos_train_sampled_10.pkl
+              - waymo_processed_data_v0_5_0_waymo_dbinfos_train_sampled_1.pkl
+
+          USE_SHARED_MEMORY: False  # set it to True to speed up (it costs about 15GB shared memory)
+          DB_DATA_PATH:
+              - waymo_processed_data_v0_5_0_gt_database_train_sampled_1_global.npy
+
           PREPARE: {
              filter_by_min_points: ['Vehicle:5', 'Pedestrian:5', 'Cyclist:5'],
              filter_by_difficulty: [-1],
@@ -62,6 +73,6 @@ DATA_PROCESSOR:
       VOXEL_SIZE: [0.1, 0.1, 0.15]
       MAX_POINTS_PER_VOXEL: 5
       MAX_NUMBER_OF_VOXELS: {
-        'train': 80000,
-        'test': 90000
+        'train': 150000,
+        'test': 150000
       }

tools/cfgs/waymo_models/PartA2.yaml

@@ -119,9 +119,12 @@ MODEL:
             TEST:
                 NMS_TYPE: nms_gpu
                 MULTI_CLASSES_NMS: False
-                NMS_PRE_MAXSIZE: 1024
-                NMS_POST_MAXSIZE: 100
-                NMS_THRESH: 0.7
+#                NMS_PRE_MAXSIZE: 1024
+#                NMS_POST_MAXSIZE: 100
+#                NMS_THRESH: 0.7
+                NMS_PRE_MAXSIZE: 4096
+                NMS_POST_MAXSIZE: 300
+                NMS_THRESH: 0.85
 
         ROI_AWARE_POOL:
             POOL_SIZE: 10
@@ -164,13 +167,13 @@ MODEL:
         NMS_CONFIG:
             MULTI_CLASSES_NMS: False
             NMS_TYPE: nms_gpu
-            NMS_THRESH: 0.1
+            NMS_THRESH: 0.7
             NMS_PRE_MAXSIZE: 4096
             NMS_POST_MAXSIZE: 500
 
 
 OPTIMIZATION:
-    BATCH_SIZE_PER_GPU: 3
+    BATCH_SIZE_PER_GPU: 2
     NUM_EPOCHS: 30
 
     OPTIMIZER: adam_onecycle

tools/cfgs/waymo_models/centerpoint.yaml

+CLASS_NAMES: ['Vehicle', 'Pedestrian', 'Cyclist']
+
+DATA_CONFIG:
+    _BASE_CONFIG_: cfgs/dataset_configs/waymo_dataset.yaml
+
+MODEL:
+    NAME: CenterPoint
+
+    VFE:
+        NAME: MeanVFE
+
+    BACKBONE_3D:
+        NAME: VoxelResBackBone8x
+
+    MAP_TO_BEV:
+        NAME: HeightCompression
+        NUM_BEV_FEATURES: 256
+
+    BACKBONE_2D:
+        NAME: BaseBEVBackbone
+
+        LAYER_NUMS: [5, 5]
+        LAYER_STRIDES: [1, 2]
+        NUM_FILTERS: [128, 256]
+        UPSAMPLE_STRIDES: [1, 2]
+        NUM_UPSAMPLE_FILTERS: [256, 256]
+
+    DENSE_HEAD:
+        NAME: CenterHead
+        CLASS_AGNOSTIC: False
+
+        CLASS_NAMES_EACH_HEAD: [
+            ['Vehicle', 'Pedestrian', 'Cyclist']
+        ]
+
+        SHARED_CONV_CHANNEL: 64
+        USE_BIAS_BEFORE_NORM: True
+        NUM_HM_CONV: 2
+        SEPARATE_HEAD_CFG:
+            HEAD_ORDER: ['center', 'center_z', 'dim', 'rot']
+            HEAD_DICT: {
+                'center': {'out_channels': 2, 'num_conv': 2},
+                'center_z': {'out_channels': 1, 'num_conv': 2},
+                'dim': {'out_channels': 3, 'num_conv': 2},
+                'rot': {'out_channels': 2, 'num_conv': 2},
+            }
+
+        TARGET_ASSIGNER_CONFIG:
+            FEATURE_MAP_STRIDE: 8
+            NUM_MAX_OBJS: 500
+            GAUSSIAN_OVERLAP: 0.1
+            MIN_RADIUS: 2
+
+        LOSS_CONFIG:
+            LOSS_WEIGHTS: {
+                'cls_weight': 1.0,
+                'loc_weight': 2.0,
+                'code_weights': [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
+            }
+
+        POST_PROCESSING:
+            SCORE_THRESH: 0.1
+            POST_CENTER_LIMIT_RANGE: [-75.2, -75.2, -2, 75.2, 75.2, 4]
+            MAX_OBJ_PER_SAMPLE: 500
+            NMS_CONFIG:
+                NMS_TYPE: nms_gpu
+                NMS_THRESH: 0.7
+                NMS_PRE_MAXSIZE: 4096
+                NMS_POST_MAXSIZE: 500
+
+    POST_PROCESSING:
+        RECALL_THRESH_LIST: [0.3, 0.5, 0.7]
+
+        EVAL_METRIC: waymo
+
+
+OPTIMIZATION:
+    BATCH_SIZE_PER_GPU: 4
+    NUM_EPOCHS: 30
+
+    OPTIMIZER: adam_onecycle
+    LR: 0.003
+    WEIGHT_DECAY: 0.01
+    MOMENTUM: 0.9
+
+    MOMS: [0.95, 0.85]
+    PCT_START: 0.4
+    DIV_FACTOR: 10
+    DECAY_STEP_LIST: [35, 45]
+    LR_DECAY: 0.1
+    LR_CLIP: 0.0000001
+
+    LR_WARMUP: False
+    WARMUP_EPOCH: 1
+
+    GRAD_NORM_CLIP: 10

tools/cfgs/waymo_models/centerpoint_without_resnet.yaml

+CLASS_NAMES: ['Vehicle', 'Pedestrian', 'Cyclist']
+
+DATA_CONFIG:
+    _BASE_CONFIG_: cfgs/dataset_configs/waymo_dataset.yaml
+
+MODEL:
+    NAME: CenterPoint
+
+    VFE:
+        NAME: MeanVFE
+
+    BACKBONE_3D:
+        NAME: VoxelBackBone8x
+
+    MAP_TO_BEV:
+        NAME: HeightCompression
+        NUM_BEV_FEATURES: 256
+
+    BACKBONE_2D:
+        NAME: BaseBEVBackbone
+
+        LAYER_NUMS: [5, 5]
+        LAYER_STRIDES: [1, 2]
+        NUM_FILTERS: [128, 256]
+        UPSAMPLE_STRIDES: [1, 2]
+        NUM_UPSAMPLE_FILTERS: [256, 256]
+
+    DENSE_HEAD:
+        NAME: CenterHead
+        CLASS_AGNOSTIC: False
+
+        CLASS_NAMES_EACH_HEAD: [
+            ['Vehicle', 'Pedestrian', 'Cyclist']
+        ]
+
+        SHARED_CONV_CHANNEL: 64
+        USE_BIAS_BEFORE_NORM: True
+        NUM_HM_CONV: 2
+        SEPARATE_HEAD_CFG:
+            HEAD_ORDER: ['center', 'center_z', 'dim', 'rot']
+            HEAD_DICT: {
+                'center': {'out_channels': 2, 'num_conv': 2},
+                'center_z': {'out_channels': 1, 'num_conv': 2},
+                'dim': {'out_channels': 3, 'num_conv': 2},
+                'rot': {'out_channels': 2, 'num_conv': 2},
+            }
+
+        TARGET_ASSIGNER_CONFIG:
+            FEATURE_MAP_STRIDE: 8
+            NUM_MAX_OBJS: 500
+            GAUSSIAN_OVERLAP: 0.1
+            MIN_RADIUS: 2
+
+        LOSS_CONFIG:
+            LOSS_WEIGHTS: {
+                'cls_weight': 1.0,
+                'loc_weight': 2.0,
+                'code_weights': [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
+            }
+
+        POST_PROCESSING:
+            SCORE_THRESH: 0.1
+            POST_CENTER_LIMIT_RANGE: [-75.2, -75.2, -2, 75.2, 75.2, 4]
+            MAX_OBJ_PER_SAMPLE: 500
+            NMS_CONFIG:
+                NMS_TYPE: nms_gpu
+                NMS_THRESH: 0.7
+                NMS_PRE_MAXSIZE: 4096
+                NMS_POST_MAXSIZE: 500
+
+    POST_PROCESSING:
+        RECALL_THRESH_LIST: [0.3, 0.5, 0.7]
+
+        EVAL_METRIC: waymo
+
+
+OPTIMIZATION:
+    BATCH_SIZE_PER_GPU: 4
+    NUM_EPOCHS: 30
+
+    OPTIMIZER: adam_onecycle
+    LR: 0.003
+    WEIGHT_DECAY: 0.01
+    MOMENTUM: 0.9
+
+    MOMS: [0.95, 0.85]
+    PCT_START: 0.4
+    DIV_FACTOR: 10
+    DECAY_STEP_LIST: [35, 45]
+    LR_DECAY: 0.1
+    LR_CLIP: 0.0000001
+
+    LR_WARMUP: False
+    WARMUP_EPOCH: 1
+
+    GRAD_NORM_CLIP: 10

tools/cfgs/waymo_models/pv_rcnn.yaml

@@ -150,9 +150,13 @@ MODEL:
             TEST:
                 NMS_TYPE: nms_gpu
                 MULTI_CLASSES_NMS: False
-                NMS_PRE_MAXSIZE: 1024
-                NMS_POST_MAXSIZE: 100
-                NMS_THRESH: 0.7
+#                NMS_PRE_MAXSIZE: 1024
+#                NMS_POST_MAXSIZE: 100
+#                NMS_THRESH: 0.7
+                NMS_PRE_MAXSIZE: 4096
+                NMS_POST_MAXSIZE: 300
+                NMS_THRESH: 0.85
+
 
         ROI_GRID_POOL:
             GRID_SIZE: 6
@@ -197,7 +201,7 @@ MODEL:
         NMS_CONFIG:
             MULTI_CLASSES_NMS: False
             NMS_TYPE: nms_gpu
-            NMS_THRESH: 0.1
+            NMS_THRESH: 0.7
             NMS_PRE_MAXSIZE: 4096
             NMS_POST_MAXSIZE: 500
 

tools/cfgs/waymo_models/pv_rcnn_with_centerhead_rpn.yaml

+CLASS_NAMES: ['Vehicle', 'Pedestrian', 'Cyclist']
+
+DATA_CONFIG:
+    _BASE_CONFIG_: cfgs/dataset_configs/waymo_dataset.yaml
+
+
+MODEL:
+    NAME: PVRCNN
+
+    VFE:
+        NAME: MeanVFE
+
+    BACKBONE_3D:
+        NAME: VoxelBackBone8x
+
+    MAP_TO_BEV:
+        NAME: HeightCompression
+        NUM_BEV_FEATURES: 256
+
+    BACKBONE_2D:
+        NAME: BaseBEVBackbone
+
+        LAYER_NUMS: [5, 5]
+        LAYER_STRIDES: [1, 2]
+        NUM_FILTERS: [128, 256]
+        UPSAMPLE_STRIDES: [1, 2]
+        NUM_UPSAMPLE_FILTERS: [256, 256]
+
+    DENSE_HEAD:
+        NAME: CenterHead
+        CLASS_AGNOSTIC: False
+
+        CLASS_NAMES_EACH_HEAD: [
+            [ 'Vehicle', 'Pedestrian', 'Cyclist' ]
+        ]
+
+        SHARED_CONV_CHANNEL: 64
+        USE_BIAS_BEFORE_NORM: True
+        NUM_HM_CONV: 2
+        SEPARATE_HEAD_CFG:
+            HEAD_ORDER: [ 'center', 'center_z', 'dim', 'rot' ]
+            HEAD_DICT: {
+                'center': { 'out_channels': 2, 'num_conv': 2 },
+                'center_z': { 'out_channels': 1, 'num_conv': 2 },
+                'dim': { 'out_channels': 3, 'num_conv': 2 },
+                'rot': { 'out_channels': 2, 'num_conv': 2 },
+            }
+
+        TARGET_ASSIGNER_CONFIG:
+            FEATURE_MAP_STRIDE: 8
+            NUM_MAX_OBJS: 500
+            GAUSSIAN_OVERLAP: 0.1
+            MIN_RADIUS: 2
+
+        LOSS_CONFIG:
+            LOSS_WEIGHTS: {
+                'cls_weight': 1.0,
+                'loc_weight': 2.0,
+                'code_weights': [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ]
+            }
+
+        POST_PROCESSING:
+            SCORE_THRESH: 0.1
+            POST_CENTER_LIMIT_RANGE: [ -75.2, -75.2, -2, 75.2, 75.2, 4 ]
+            MAX_OBJ_PER_SAMPLE: 500
+            NMS_CONFIG:
+                NMS_TYPE: nms_gpu
+                NMS_THRESH: 0.7
+                NMS_PRE_MAXSIZE: 4096
+                NMS_POST_MAXSIZE: 500
+
+    PFE:
+        NAME: VoxelSetAbstraction
+        POINT_SOURCE: raw_points
+        NUM_KEYPOINTS: 4096
+        NUM_OUTPUT_FEATURES: 90
+        SAMPLE_METHOD: FPS
+
+        FEATURES_SOURCE: ['bev', 'x_conv3', 'x_conv4', 'raw_points']
+        SA_LAYER:
+            raw_points:
+                MLPS: [[16, 16], [16, 16]]
+                POOL_RADIUS: [0.4, 0.8]
+                NSAMPLE: [16, 16]
+            x_conv1:
+                DOWNSAMPLE_FACTOR: 1
+                MLPS: [[16, 16], [16, 16]]
+                POOL_RADIUS: [0.4, 0.8]
+                NSAMPLE: [16, 16]
+            x_conv2:
+                DOWNSAMPLE_FACTOR: 2
+                MLPS: [[32, 32], [32, 32]]
+                POOL_RADIUS: [0.8, 1.2]
+                NSAMPLE: [16, 32]
+            x_conv3:
+                DOWNSAMPLE_FACTOR: 4
+                MLPS: [[64, 64], [64, 64]]
+                POOL_RADIUS: [1.2, 2.4]
+                NSAMPLE: [16, 32]
+            x_conv4:
+                DOWNSAMPLE_FACTOR: 8
+                MLPS: [[64, 64], [64, 64]]
+                POOL_RADIUS: [2.4, 4.8]
+                NSAMPLE: [16, 32]
+
+    POINT_HEAD:
+        NAME: PointHeadSimple
+        CLS_FC: [256, 256]
+        CLASS_AGNOSTIC: True
+        USE_POINT_FEATURES_BEFORE_FUSION: True
+        TARGET_CONFIG:
+            GT_EXTRA_WIDTH: [0.2, 0.2, 0.2]
+        LOSS_CONFIG:
+            LOSS_REG: smooth-l1
+            LOSS_WEIGHTS: {
+                'point_cls_weight': 1.0,
+            }
+
+    ROI_HEAD:
+        NAME: PVRCNNHead
+        CLASS_AGNOSTIC: True
+
+        SHARED_FC: [256, 256]
+        CLS_FC: [256, 256]
+        REG_FC: [256, 256]
+        DP_RATIO: 0.3
+
+        NMS_CONFIG:
+            TRAIN:
+                NMS_TYPE: nms_gpu
+                MULTI_CLASSES_NMS: False
+                NMS_PRE_MAXSIZE: 9000
+                NMS_POST_MAXSIZE: 512
+                NMS_THRESH: 0.8
+            TEST:
+                NMS_TYPE: nms_gpu
+                MULTI_CLASSES_NMS: False
+                NMS_PRE_MAXSIZE: 1024
+                NMS_POST_MAXSIZE: 100
+                NMS_THRESH: 0.7
+                SCORE_THRESH: 0.1
+
+#                NMS_PRE_MAXSIZE: 4096
+#                NMS_POST_MAXSIZE: 500
+#                NMS_THRESH: 0.85
+
+
+        ROI_GRID_POOL:
+            GRID_SIZE: 6
+            MLPS: [[64, 64], [64, 64]]
+            POOL_RADIUS: [0.8, 1.6]
+            NSAMPLE: [16, 16]
+            POOL_METHOD: max_pool
+
+        TARGET_CONFIG:
+            BOX_CODER: ResidualCoder
+            ROI_PER_IMAGE: 128
+            FG_RATIO: 0.5
+
+            SAMPLE_ROI_BY_EACH_CLASS: True
+            CLS_SCORE_TYPE: roi_iou
+
+            CLS_FG_THRESH: 0.75
+            CLS_BG_THRESH: 0.25
+            CLS_BG_THRESH_LO: 0.1
+            HARD_BG_RATIO: 0.8
+
+            REG_FG_THRESH: 0.55
+
+        LOSS_CONFIG:
+            CLS_LOSS: BinaryCrossEntropy
+            REG_LOSS: smooth-l1
+            CORNER_LOSS_REGULARIZATION: True
+            LOSS_WEIGHTS: {
+                'rcnn_cls_weight': 1.0,
+                'rcnn_reg_weight': 1.0,
+                'rcnn_corner_weight': 1.0,
+                'code_weights': [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
+            }
+
+    POST_PROCESSING:
+        RECALL_THRESH_LIST: [0.3, 0.5, 0.7]
+        SCORE_THRESH: 0.1
+        OUTPUT_RAW_SCORE: False
+
+        EVAL_METRIC: waymo
+
+        NMS_CONFIG:
+            MULTI_CLASSES_NMS: False
+            NMS_TYPE: nms_gpu
+            NMS_THRESH: 0.7
+            NMS_PRE_MAXSIZE: 4096
+            NMS_POST_MAXSIZE: 500
+
+
+OPTIMIZATION:
+    BATCH_SIZE_PER_GPU: 2
+    NUM_EPOCHS: 30
+
+    OPTIMIZER: adam_onecycle
+    LR: 0.01
+    WEIGHT_DECAY: 0.001
+    MOMENTUM: 0.9
+
+    MOMS: [0.95, 0.85]
+    PCT_START: 0.4
+    DIV_FACTOR: 10
+    DECAY_STEP_LIST: [35, 45]
+    LR_DECAY: 0.1
+    LR_CLIP: 0.0000001
+
+    LR_WARMUP: False
+    WARMUP_EPOCH: 1
+
+    GRAD_NORM_CLIP: 10
\ No newline at end of file

tools/cfgs/waymo_models/second.yaml

@@ -94,7 +94,7 @@ MODEL:
         NMS_CONFIG:
             MULTI_CLASSES_NMS: False
             NMS_TYPE: nms_gpu
-            NMS_THRESH: 0.01
+            NMS_THRESH: 0.7
             NMS_PRE_MAXSIZE: 4096
             NMS_POST_MAXSIZE: 500
 

tools/demo.py

@@ -2,7 +2,15 @@ import argparse
 import glob
 from pathlib import Path
 
-import mayavi.mlab as mlab
+try:
+    import open3d
+    from visual_utils import open3d_vis_utils as V
+    OPEN3D_FLAG = True
+except:
+    import mayavi.mlab as mlab
+    from visual_utils import visualize_utils as V
+    OPEN3D_FLAG = False
+
 import numpy as np
 import torch
 
@@ -10,7 +18,6 @@ from pcdet.config import cfg, cfg_from_yaml_file
 from pcdet.datasets import DatasetTemplate
 from pcdet.models import build_network, load_data_to_gpu
 from pcdet.utils import common_utils
-from visual_utils import visualize_utils as V
 
 
 class DemoDataset(DatasetTemplate):
@@ -94,6 +101,8 @@ def main():
                 points=data_dict['points'][:, 1:], ref_boxes=pred_dicts[0]['pred_boxes'],
                 ref_scores=pred_dicts[0]['pred_scores'], ref_labels=pred_dicts[0]['pred_labels']
             )
+
+            if not OPEN3D_FLAG:
                 mlab.show(stop=True)
 
     logger.info('Demo done.')

tools/test.py

+import _init_path
 import argparse
 import datetime
 import glob

tools/train.py

+import _init_path
 import argparse
 import datetime
 import glob
@@ -6,7 +7,6 @@ from pathlib import Path
 from test import repeat_eval_ckpt
 
 import torch
-import torch.distributed as dist
 import torch.nn as nn
 from tensorboardX import SummaryWriter
 
@@ -24,7 +24,7 @@ def parse_config():
 
     parser.add_argument('--batch_size', type=int, default=None, required=False, help='batch size for training')
     parser.add_argument('--epochs', type=int, default=None, required=False, help='number of epochs to train for')
-    parser.add_argument('--workers', type=int, default=8, help='number of workers for dataloader')
+    parser.add_argument('--workers', type=int, default=4, help='number of workers for dataloader')
     parser.add_argument('--extra_tag', type=str, default='default', help='extra tag for this experiment')
     parser.add_argument('--ckpt', type=str, default=None, help='checkpoint to start from')
     parser.add_argument('--pretrained_model', type=str, default=None, help='pretrained_model')
@@ -170,6 +170,9 @@ def main():
         merge_all_iters_to_one_epoch=args.merge_all_iters_to_one_epoch
     )
 
+    if hasattr(train_set, 'use_shared_memory') and train_set.use_shared_memory:
+        train_set.clean_shared_memory()
+
     logger.info('**********************End training %s/%s(%s)**********************\n\n\n'
                 % (cfg.EXP_GROUP_PATH, cfg.TAG, args.extra_tag))
 
@@ -183,7 +186,7 @@ def main():
     )
     eval_output_dir = output_dir / 'eval' / 'eval_with_train'
     eval_output_dir.mkdir(parents=True, exist_ok=True)
-    args.start_epoch = max(args.epochs - 10, 0)  # Only evaluate the last 10 epochs
+    args.start_epoch = max(args.epochs - 0, 0)  # Only evaluate the last 10 epochs
 
     repeat_eval_ckpt(
         model.module if dist_train else model,

tools/train_utils/optimization/fastai_optim.py

 # This file is modified from https://github.com/traveller59/second.pytorch
 
-from collections import Iterable
+try:
+    from collections.abc import Iterable
+except:
+    from collections import Iterable
 
 import torch
 from torch import nn

tools/visual_utils/open3d_vis_utils.py

+"""
+Open3d visualization tool box
+Written by Jihan YANG
+All rights preserved from 2021 - present.
+"""
+import open3d
+import torch
+import matplotlib
+import numpy as np
+
+box_colormap = [
+    [1, 1, 1],
+    [0, 1, 0],
+    [0, 1, 1],
+    [1, 1, 0],
+]
+
+
+def get_coor_colors(obj_labels):
+    """
+    Args:
+        obj_labels: 1 is ground, labels > 1 indicates different instance cluster
+
+    Returns:
+        rgb: [N, 3]. color for each point.
+    """
+    colors = matplotlib.colors.XKCD_COLORS.values()
+    max_color_num = obj_labels.max()
+
+    color_list = list(colors)[:max_color_num+1]
+    colors_rgba = [matplotlib.colors.to_rgba_array(color) for color in color_list]
+    label_rgba = np.array(colors_rgba)[obj_labels]
+    label_rgba = label_rgba.squeeze()[:, :3]
+
+    return label_rgba
+
+
+def draw_scenes(points, gt_boxes=None, ref_boxes=None, ref_labels=None, ref_scores=None, point_colors=None, draw_origin=True):
+    if isinstance(points, torch.Tensor):
+        points = points.cpu().numpy()
+    if isinstance(gt_boxes, torch.Tensor):
+        gt_boxes = gt_boxes.cpu().numpy()
+    if isinstance(ref_boxes, torch.Tensor):
+        ref_boxes = ref_boxes.cpu().numpy()
+
+    vis = open3d.visualization.Visualizer()
+    vis.create_window()
+
+    vis.get_render_option().point_size = 1.0
+    vis.get_render_option().background_color = np.zeros(3)
+
+    # draw origin
+    if draw_origin:
+        axis_pcd = open3d.geometry.TriangleMesh.create_coordinate_frame(size=1.0, origin=[0, 0, 0])
+        vis.add_geometry(axis_pcd)
+
+    pts = open3d.geometry.PointCloud()
+    pts.points = open3d.utility.Vector3dVector(points[:, :3])
+
+    vis.add_geometry(pts)
+    if point_colors is None:
+        pts.colors = open3d.utility.Vector3dVector(np.ones((points.shape[0], 3)))
+    else:
+        pts.colors = open3d.utility.Vector3dVector(point_colors)
+
+    if gt_boxes is not None:
+        vis = draw_box(vis, gt_boxes, (0, 0, 1))
+
+    if ref_boxes is not None:
+        vis = draw_box(vis, ref_boxes, (0, 1, 0), ref_labels, ref_scores)
+
+    vis.run()
+    vis.destroy_window()
+
+
+def translate_boxes_to_open3d_instance(gt_boxes):
+    """
+             4-------- 6
+           /|         /|
+          5 -------- 3 .
+          | |        | |
+          . 7 -------- 1
+          |/         |/
+          2 -------- 0
+    """
+    center = gt_boxes[0:3]
+    lwh = gt_boxes[3:6]
+    axis_angles = np.array([0, 0, gt_boxes[6] + 1e-10])
+    rot = open3d.geometry.get_rotation_matrix_from_axis_angle(axis_angles)
+    box3d = open3d.geometry.OrientedBoundingBox(center, rot, lwh)
+
+    line_set = open3d.geometry.LineSet.create_from_oriented_bounding_box(box3d)
+
+    # import ipdb; ipdb.set_trace(context=20)
+    lines = np.asarray(line_set.lines)
+    lines = np.concatenate([lines, np.array([[1, 4], [7, 6]])], axis=0)
+
+    line_set.lines = open3d.utility.Vector2iVector(lines)
+
+    return line_set, box3d
+
+
+def draw_box(vis, gt_boxes, color=(0, 1, 0), ref_labels=None, score=None):
+    for i in range(gt_boxes.shape[0]):
+        line_set, box3d = translate_boxes_to_open3d_instance(gt_boxes[i])
+        if ref_labels is None:
+            line_set.paint_uniform_color(color)
+        else:
+            line_set.paint_uniform_color(box_colormap[ref_labels[i]])
+
+        vis.add_geometry(line_set)
+
+        # if score is not None:
+        #     corners = box3d.get_box_points()
+        #     vis.add_3d_label(corners[5], '%.2f' % score[i])
+    return vis