Closed Rrrengar closed 1 year ago
yukang2017
commented
1 year ago Hi,
I have implemented a version of LargeKernel3D on OpenPCDet. But it is LIDAR-only, without multi-modal settings. Because OpenPCDet does not support multi modality currently. Do you still need it? If yes, I will release it this week.
Regards, Yukang Chen
Rrrengar
commented
1 year ago Hi,
I have implemented a version of LargeKernel3D on OpenPCDet. But it is LIDAR-only, without multi-modal settings. Because OpenPCDet does not support multi modality currently. Do you still need it? If yes, I will release it this week.
Regards, Yukang Chen
Hi! Thanks for your reply and the LIDAR-only version on OpenPCDet is just the one I need. That would be helpful if you release it this week!
jianingwangind
commented
1 year ago Also needed the OpenPCDet version, looking forward to your code release.
yukang2017
commented
1 year ago Hi,
I have uploaded it here.
Regards, Yukang Chen
jianingwangind
commented
1 year ago @yukang2017 Thanks a lot. Could you also share the config file maybe the pretrained model as well? Thousand thanks!
yukang2017
commented
1 year ago Hi @jianingwangind ,
The config file is very simple. You can simply change
NAME: VoxelResBackBone8x to NAME: VoxelResBackBone8xLargeKernel3D in
https://github.com/open-mmlab/OpenPCDet/blob/ad9c25c03cf9e866bbee8c48fa4e2860cce68a9c/tools/cfgs/nuscenes_models/cbgs_voxel0075_res3d_centerpoint.yaml#L70
The pre-trained model can be downloaded via the link below here. Note that spatialgroupconvv2 is used in this pre-trained weight.
https://drive.google.com/file/d/1ZiWWxtkGIqelj9dG-7SJ2pMAJIwVVP1l/view?usp=share_link
This is the performance on validation set for this pre-trained weight.
***car error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.18, 0.15, 0.14, 0.22, 0.18 | 75.83, 85.54, 88.76, 90.45 | mean AP: 0.8514791433295339
***truck error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.32, 0.18, 0.15, 0.19, 0.21 | 42.01, 59.46, 67.54, 71.40 | mean AP: 0.6010389384423573
***construction_vehicle error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.72, 0.42, 0.91, 0.13, 0.31 | 3.48, 14.76, 33.14, 45.94 | mean AP: 0.24330653887851947
***bus error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.31, 0.19, 0.07, 0.34, 0.25 | 49.23, 72.38, 82.99, 85.63 | mean AP: 0.7256052094405145
***trailer error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.51, 0.19, 0.50, 0.17, 0.18 | 10.57, 37.94, 52.72, 64.26 | mean AP: 0.4137271105906273
***barrier error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.19, 0.28, 0.12, nan, nan | 59.12, 67.77, 71.29, 72.75 | mean AP: 0.6773148284011004
***motorcycle error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.20, 0.23, 0.33, 0.29, 0.25 | 61.46, 72.70, 74.30, 74.63 | mean AP: 0.7077321358715435
***bicycle error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.16, 0.26, 0.51, 0.15, 0.01 | 57.00, 59.43, 59.96, 60.44 | mean AP: 0.5920782509164128
***pedestrian error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.14, 0.27, 0.41, 0.20, 0.09 | 83.58, 85.13, 86.18, 87.57 | mean AP: 0.856159033016184
***traffic_cone error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.13, 0.32, nan, nan, nan | 69.47, 71.03, 72.76, 75.76 | mean AP: 0.722540472275063
--------------average performance-------------
trans_err: 0.2855
scale_err: 0.2501
orient_err: 0.3506
vel_err: 0.2111
attr_err: 0.1871
mAP: 0.6391
NDS: 0.6911Regards, Yukang Chen
jianingwangind
commented
1 year ago @yukang2017 I really appreciate for your detailed explanation, but unfortunately i encountered this error, do you have any ideas?
Looking forward to your reply
yukang2017
commented
1 year ago Hi @jianingwangind ,
I have fixed it via this line. Would you please try it again? https://github.com/dvlab-research/LargeKernel3D/blob/0620e8d57c73652744001592bef909837d94c630/object-detection/pcdet/models/backbones_3d/spconv_backbone_largekernel.py#L230
Regards, Yukang Chen
jianingwangind
commented
1 year ago @yukang2017 Hi, thanks for your reply. It worked, but the metrics seems lower than yours. Do you have any ideas? Thanks and looking forward to your reply
yukang2017
commented
1 year ago Hi,
Would you please show me your complete test log for me?
Regards, Yukang Chen
jianingwangind
commented
1 year ago Hi @yukang2017 , thanks for your reply and the complete test log is listed below:
ssh://root/opt/conda/bin/python -u /jacob.wang/projects/OpenPCDet/tools/test.py --cfg_file tools/cfgs/nuscenes_models/cbgs_voxel0075_largekernel_centerpoint.yaml --batch_size 1 --ckpt ckpts/largekernel3D_centerpoint_openpcdet.pth
2023-05-09 03:23:24,510 INFO **********************Start logging**********************
2023-05-09 03:23:24,510 INFO CUDA_VISIBLE_DEVICES=ALL
2023-05-09 03:23:24,510 INFO cfg_file tools/cfgs/nuscenes_models/cbgs_voxel0075_largekernel_centerpoint.yaml
2023-05-09 03:23:24,511 INFO batch_size 1
2023-05-09 03:23:24,511 INFO workers 4
2023-05-09 03:23:24,511 INFO extra_tag default
2023-05-09 03:23:24,511 INFO ckpt ckpts/largekernel3D_centerpoint_openpcdet.pth
2023-05-09 03:23:24,511 INFO pretrained_model None
2023-05-09 03:23:24,511 INFO launcher none
2023-05-09 03:23:24,512 INFO tcp_port 18888
2023-05-09 03:23:24,512 INFO local_rank 0
2023-05-09 03:23:24,512 INFO set_cfgs None
2023-05-09 03:23:24,512 INFO max_waiting_mins 30
2023-05-09 03:23:24,512 INFO start_epoch 0
2023-05-09 03:23:24,512 INFO eval_tag default
2023-05-09 03:23:24,513 INFO eval_all False
2023-05-09 03:23:24,513 INFO ckpt_dir None
2023-05-09 03:23:24,513 INFO save_to_file False
2023-05-09 03:23:24,513 INFO infer_time False
2023-05-09 03:23:24,513 INFO cfg.ROOT_DIR: /jacob.wang/projects/OpenPCDet
2023-05-09 03:23:24,513 INFO cfg.LOCAL_RANK: 0
2023-05-09 03:23:24,513 INFO cfg.CLASS_NAMES: ['car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier', 'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone']
2023-05-09 03:23:24,514 INFO ----------- DATA_CONFIG -----------
2023-05-09 03:23:24,514 INFO cfg.DATA_CONFIG.DATASET: NuScenesDataset
2023-05-09 03:23:24,514 INFO cfg.DATA_CONFIG.DATA_PATH: data/nuscenes
2023-05-09 03:23:24,514 INFO cfg.DATA_CONFIG.VERSION: v1.0-trainval
2023-05-09 03:23:24,514 INFO cfg.DATA_CONFIG.MAX_SWEEPS: 10
2023-05-09 03:23:24,514 INFO cfg.DATA_CONFIG.PRED_VELOCITY: True
2023-05-09 03:23:24,515 INFO cfg.DATA_CONFIG.SET_NAN_VELOCITY_TO_ZEROS: True
2023-05-09 03:23:24,515 INFO cfg.DATA_CONFIG.FILTER_MIN_POINTS_IN_GT: 1
2023-05-09 03:23:24,515 INFO ----------- DATA_SPLIT -----------
2023-05-09 03:23:24,515 INFO cfg.DATA_CONFIG.DATA_SPLIT.train: train
2023-05-09 03:23:24,515 INFO cfg.DATA_CONFIG.DATA_SPLIT.test: val
2023-05-09 03:23:24,515 INFO ----------- INFO_PATH -----------
2023-05-09 03:23:24,515 INFO cfg.DATA_CONFIG.INFO_PATH.train: ['nuscenes_infos_10sweeps_train.pkl']
2023-05-09 03:23:24,516 INFO cfg.DATA_CONFIG.INFO_PATH.test: ['nuscenes_infos_10sweeps_val.pkl']
2023-05-09 03:23:24,516 INFO cfg.DATA_CONFIG.POINT_CLOUD_RANGE: [-54.0, -54.0, -5.0, 54.0, 54.0, 3.0]
2023-05-09 03:23:24,516 INFO cfg.DATA_CONFIG.BALANCED_RESAMPLING: True
2023-05-09 03:23:24,516 INFO ----------- DATA_AUGMENTOR -----------
2023-05-09 03:23:24,516 INFO cfg.DATA_CONFIG.DATA_AUGMENTOR.DISABLE_AUG_LIST: ['placeholder']
2023-05-09 03:23:24,516 INFO cfg.DATA_CONFIG.DATA_AUGMENTOR.AUG_CONFIG_LIST: [{'NAME': 'gt_sampling', 'DB_INFO_PATH': ['nuscenes_dbinfos_10sweeps_withvelo.pkl'], 'PREPARE': {'filter_by_min_points': ['car:5', 'truck:5', 'construction_vehicle:5', 'bus:5', 'trailer:5', 'barrier:5', 'motorcycle:5', 'bicycle:5', 'pedestrian:5', 'traffic_cone:5']}, 'SAMPLE_GROUPS': ['car:2', 'truck:3', 'construction_vehicle:7', 'bus:4', 'trailer:6', 'barrier:2', 'motorcycle:6', 'bicycle:6', 'pedestrian:2', 'traffic_cone:2'], 'NUM_POINT_FEATURES': 5, 'DATABASE_WITH_FAKELIDAR': False, 'REMOVE_EXTRA_WIDTH': [0.0, 0.0, 0.0], 'LIMIT_WHOLE_SCENE': True}, {'NAME': 'random_world_flip', 'ALONG_AXIS_LIST': ['x', 'y']}, {'NAME': 'random_world_rotation', 'WORLD_ROT_ANGLE': [-0.78539816, 0.78539816]}, {'NAME': 'random_world_scaling', 'WORLD_SCALE_RANGE': [0.9, 1.1]}, {'NAME': 'random_world_translation', 'NOISE_TRANSLATE_STD': [0.5, 0.5, 0.5]}]
2023-05-09 03:23:24,516 INFO ----------- POINT_FEATURE_ENCODING -----------
2023-05-09 03:23:24,516 INFO cfg.DATA_CONFIG.POINT_FEATURE_ENCODING.encoding_type: absolute_coordinates_encoding
2023-05-09 03:23:24,516 INFO cfg.DATA_CONFIG.POINT_FEATURE_ENCODING.used_feature_list: ['x', 'y', 'z', 'intensity', 'timestamp']
2023-05-09 03:23:24,516 INFO cfg.DATA_CONFIG.POINT_FEATURE_ENCODING.src_feature_list: ['x', 'y', 'z', 'intensity', 'timestamp']
2023-05-09 03:23:24,516 INFO cfg.DATA_CONFIG.DATA_PROCESSOR: [{'NAME': 'mask_points_and_boxes_outside_range', 'REMOVE_OUTSIDE_BOXES': True}, {'NAME': 'shuffle_points', 'SHUFFLE_ENABLED': {'train': True, 'test': True}}, {'NAME': 'transform_points_to_voxels', 'VOXEL_SIZE': [0.075, 0.075, 0.2], 'MAX_POINTS_PER_VOXEL': 10, 'MAX_NUMBER_OF_VOXELS': {'train': 120000, 'test': 160000}}]
2023-05-09 03:23:24,517 INFO cfg.DATA_CONFIG._BASE_CONFIG_: tools/cfgs/dataset_configs/nuscenes_dataset.yaml
2023-05-09 03:23:24,517 INFO ----------- MODEL -----------
2023-05-09 03:23:24,517 INFO cfg.MODEL.NAME: CenterPoint
2023-05-09 03:23:24,517 INFO ----------- VFE -----------
2023-05-09 03:23:24,517 INFO cfg.MODEL.VFE.NAME: MeanVFE
2023-05-09 03:23:24,517 INFO ----------- BACKBONE_3D -----------
2023-05-09 03:23:24,517 INFO cfg.MODEL.BACKBONE_3D.NAME: VoxelResBackBone8xLargeKernel3D
2023-05-09 03:23:24,517 INFO ----------- MAP_TO_BEV -----------
2023-05-09 03:23:24,517 INFO cfg.MODEL.MAP_TO_BEV.NAME: HeightCompression
2023-05-09 03:23:24,517 INFO cfg.MODEL.MAP_TO_BEV.NUM_BEV_FEATURES: 256
2023-05-09 03:23:24,517 INFO ----------- BACKBONE_2D -----------
2023-05-09 03:23:24,517 INFO cfg.MODEL.BACKBONE_2D.NAME: BaseBEVBackbone
2023-05-09 03:23:24,517 INFO cfg.MODEL.BACKBONE_2D.LAYER_NUMS: [5, 5]
2023-05-09 03:23:24,517 INFO cfg.MODEL.BACKBONE_2D.LAYER_STRIDES: [1, 2]
2023-05-09 03:23:24,517 INFO cfg.MODEL.BACKBONE_2D.NUM_FILTERS: [128, 256]
2023-05-09 03:23:24,517 INFO cfg.MODEL.BACKBONE_2D.UPSAMPLE_STRIDES: [1, 2]
2023-05-09 03:23:24,517 INFO cfg.MODEL.BACKBONE_2D.NUM_UPSAMPLE_FILTERS: [256, 256]
2023-05-09 03:23:24,517 INFO ----------- DENSE_HEAD -----------
2023-05-09 03:23:24,517 INFO cfg.MODEL.DENSE_HEAD.NAME: CenterHead
2023-05-09 03:23:24,517 INFO cfg.MODEL.DENSE_HEAD.CLASS_AGNOSTIC: False
2023-05-09 03:23:24,517 INFO cfg.MODEL.DENSE_HEAD.CLASS_NAMES_EACH_HEAD: [['car'], ['truck', 'construction_vehicle'], ['bus', 'trailer'], ['barrier'], ['motorcycle', 'bicycle'], ['pedestrian', 'traffic_cone']]
2023-05-09 03:23:24,517 INFO cfg.MODEL.DENSE_HEAD.SHARED_CONV_CHANNEL: 64
2023-05-09 03:23:24,517 INFO cfg.MODEL.DENSE_HEAD.USE_BIAS_BEFORE_NORM: True
2023-05-09 03:23:24,517 INFO cfg.MODEL.DENSE_HEAD.NUM_HM_CONV: 2
2023-05-09 03:23:24,517 INFO ----------- SEPARATE_HEAD_CFG -----------
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_ORDER: ['center', 'center_z', 'dim', 'rot', 'vel']
2023-05-09 03:23:24,518 INFO ----------- HEAD_DICT -----------
2023-05-09 03:23:24,518 INFO ----------- center -----------
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.center.out_channels: 2
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.center.num_conv: 2
2023-05-09 03:23:24,518 INFO ----------- center_z -----------
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.center_z.out_channels: 1
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.center_z.num_conv: 2
2023-05-09 03:23:24,518 INFO ----------- dim -----------
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.dim.out_channels: 3
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.dim.num_conv: 2
2023-05-09 03:23:24,518 INFO ----------- rot -----------
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.rot.out_channels: 2
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.rot.num_conv: 2
2023-05-09 03:23:24,518 INFO ----------- vel -----------
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.vel.out_channels: 2
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.SEPARATE_HEAD_CFG.HEAD_DICT.vel.num_conv: 2
2023-05-09 03:23:24,518 INFO ----------- TARGET_ASSIGNER_CONFIG -----------
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.TARGET_ASSIGNER_CONFIG.FEATURE_MAP_STRIDE: 8
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.TARGET_ASSIGNER_CONFIG.NUM_MAX_OBJS: 500
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.TARGET_ASSIGNER_CONFIG.GAUSSIAN_OVERLAP: 0.1
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.TARGET_ASSIGNER_CONFIG.MIN_RADIUS: 2
2023-05-09 03:23:24,518 INFO ----------- LOSS_CONFIG -----------
2023-05-09 03:23:24,518 INFO ----------- LOSS_WEIGHTS -----------
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.LOSS_CONFIG.LOSS_WEIGHTS.cls_weight: 1.0
2023-05-09 03:23:24,518 INFO cfg.MODEL.DENSE_HEAD.LOSS_CONFIG.LOSS_WEIGHTS.loc_weight: 0.25
2023-05-09 03:23:24,519 INFO cfg.MODEL.DENSE_HEAD.LOSS_CONFIG.LOSS_WEIGHTS.code_weights: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.2, 0.2, 1.0, 1.0]
2023-05-09 03:23:24,519 INFO ----------- POST_PROCESSING -----------
2023-05-09 03:23:24,519 INFO cfg.MODEL.DENSE_HEAD.POST_PROCESSING.SCORE_THRESH: 0.1
2023-05-09 03:23:24,519 INFO cfg.MODEL.DENSE_HEAD.POST_PROCESSING.POST_CENTER_LIMIT_RANGE: [-61.2, -61.2, -10.0, 61.2, 61.2, 10.0]
2023-05-09 03:23:24,519 INFO cfg.MODEL.DENSE_HEAD.POST_PROCESSING.MAX_OBJ_PER_SAMPLE: 500
2023-05-09 03:23:24,519 INFO ----------- NMS_CONFIG -----------
2023-05-09 03:23:24,519 INFO cfg.MODEL.DENSE_HEAD.POST_PROCESSING.NMS_CONFIG.NMS_TYPE: nms_gpu
2023-05-09 03:23:24,519 INFO cfg.MODEL.DENSE_HEAD.POST_PROCESSING.NMS_CONFIG.NMS_THRESH: 0.2
2023-05-09 03:23:24,519 INFO cfg.MODEL.DENSE_HEAD.POST_PROCESSING.NMS_CONFIG.NMS_PRE_MAXSIZE: 1000
2023-05-09 03:23:24,519 INFO cfg.MODEL.DENSE_HEAD.POST_PROCESSING.NMS_CONFIG.NMS_POST_MAXSIZE: 83
2023-05-09 03:23:24,519 INFO ----------- POST_PROCESSING -----------
2023-05-09 03:23:24,519 INFO cfg.MODEL.POST_PROCESSING.RECALL_THRESH_LIST: [0.3, 0.5, 0.7]
2023-05-09 03:23:24,519 INFO cfg.MODEL.POST_PROCESSING.EVAL_METRIC: kitti
2023-05-09 03:23:24,519 INFO ----------- OPTIMIZATION -----------
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.BATCH_SIZE_PER_GPU: 4
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.NUM_EPOCHS: 20
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.OPTIMIZER: adam_onecycle
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.LR: 0.001
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.WEIGHT_DECAY: 0.01
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.MOMENTUM: 0.9
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.MOMS: [0.95, 0.85]
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.PCT_START: 0.4
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.DIV_FACTOR: 10
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.DECAY_STEP_LIST: [35, 45]
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.LR_DECAY: 0.1
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.LR_CLIP: 1e-07
2023-05-09 03:23:24,519 INFO cfg.OPTIMIZATION.LR_WARMUP: False
2023-05-09 03:23:24,520 INFO cfg.OPTIMIZATION.WARMUP_EPOCH: 1
2023-05-09 03:23:24,520 INFO cfg.OPTIMIZATION.GRAD_NORM_CLIP: 10
2023-05-09 03:23:24,520 INFO cfg.TAG: cbgs_voxel0075_largekernel_centerpoint
2023-05-09 03:23:24,520 INFO cfg.EXP_GROUP_PATH: cfgs/nuscenes_models
2023-05-09 03:23:24,520 INFO Loading NuScenes dataset
2023-05-09 03:23:25,103 INFO Total samples for NuScenes dataset: 6019
/jacob.wang/projects/OpenPCDet/pcdet/models/backbones_3d/spconv_backbone_largekernel.py:73: UserWarning: torch.range is deprecated and will be removed in a future release because its behavior is inconsistent with Python's range builtin. Instead, use torch.arange, which produces values in [start, end).
b = torch.range(0, kernel_size**3-1, 1)[a.reshape(-1).bool()]
2023-05-09 03:23:27,843 INFO ==> Loading parameters from checkpoint ckpts/largekernel3D_centerpoint_openpcdet.pth to GPU
2023-05-09 03:23:28,636 INFO ==> Checkpoint trained from version: pcdet+0.5.2+0000000
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.0.conv1.block.weight: torch.Size([7, 7, 7, 16, 16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.0.conv1.conv3x3_1.weight: torch.Size([16, 3, 3, 3, 16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.0.conv1.conv3x3_1.bias: torch.Size([16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.0.conv2.block.weight: torch.Size([7, 7, 7, 16, 16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.0.conv2.conv3x3_1.weight: torch.Size([16, 3, 3, 3, 16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.0.conv2.conv3x3_1.bias: torch.Size([16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.1.conv1.block.weight: torch.Size([7, 7, 7, 16, 16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.1.conv1.conv3x3_1.weight: torch.Size([16, 3, 3, 3, 16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.1.conv1.conv3x3_1.bias: torch.Size([16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.1.conv2.block.weight: torch.Size([7, 7, 7, 16, 16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.1.conv2.conv3x3_1.weight: torch.Size([16, 3, 3, 3, 16])
2023-05-09 03:23:29,053 INFO Not updated weight backbone_3d.conv1.1.conv2.conv3x3_1.bias: torch.Size([16])
2023-05-09 03:23:29,053 INFO ==> Done (loaded 554/566)
2023-05-09 03:23:29,074 INFO *************** EPOCH 3 EVALUATION *****************
eval: 0%| | 0/6019 [00:00<?, ?it/s]/opt/conda/lib/python3.7/site-packages/torch/_tensor.py:575: UserWarning: floor_divide is deprecated, and will be removed in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values.
To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor'). (Triggered internally at /opt/conda/conda-bld/pytorch_1623448265233/work/aten/src/ATen/native/BinaryOps.cpp:467.)
return torch.floor_divide(self, other)
eval: 100%|█| 6019/6019 [35:02<00:00, 2.86it/s, recall_0.3=(0, 113634) / 149480
2023-05-09 03:58:31,296 INFO *************** Performance of EPOCH 3 *****************
2023-05-09 03:58:31,297 INFO Generate label finished(sec_per_example: 0.3493 second).
2023-05-09 03:58:31,297 INFO recall_roi_0.3: 0.000000
2023-05-09 03:58:31,297 INFO recall_rcnn_0.3: 0.760195
2023-05-09 03:58:31,297 INFO recall_roi_0.5: 0.000000
2023-05-09 03:58:31,297 INFO recall_rcnn_0.5: 0.543417
2023-05-09 03:58:31,298 INFO recall_roi_0.7: 0.000000
2023-05-09 03:58:31,298 INFO recall_rcnn_0.7: 0.234319
2023-05-09 03:58:31,309 INFO Average predicted number of objects(6019 samples): 74.590
======
Loading NuScenes tables for version v1.0-trainval...
Loading nuScenes-lidarseg...
32 category,
8 attribute,
4 visibility,
64386 instance,
12 sensor,
10200 calibrated_sensor,
2631083 ego_pose,
68 log,
850 scene,
34149 sample,
2631083 sample_data,
1166187 sample_annotation,
4 map,
34149 lidarseg,
Done loading in 29.800 seconds.
======
Reverse indexing ...
Done reverse indexing in 8.5 seconds.
======
2023-05-09 04:00:37,886 INFO The predictions of NuScenes have been saved to /jacob.wang/projects/OpenPCDet/output/cfgs/nuscenes_models/cbgs_voxel0075_largekernel_centerpoint/default/eval/epoch_3/val/default/final_result/data/results_nusc.json
Initializing nuScenes detection evaluation
Loaded results from /jacob.wang/projects/OpenPCDet/output/cfgs/nuscenes_models/cbgs_voxel0075_largekernel_centerpoint/default/eval/epoch_3/val/default/final_result/data/results_nusc.json. Found detections for 6019 samples.
Loading annotations for val split from nuScenes version: v1.0-trainval
100%|██████████████████████████████████████| 6019/6019 [00:10<00:00, 547.60it/s]
Loaded ground truth annotations for 6019 samples.
Filtering predictions
=> Original number of boxes: 448956
=> After distance based filtering: 315264
=> After LIDAR and RADAR points based filtering: 315264
=> After bike rack filtering: 315183
Filtering ground truth annotations
=> Original number of boxes: 187528
=> After distance based filtering: 134565
=> After LIDAR and RADAR points based filtering: 121871
=> After bike rack filtering: 121861
Accumulating metric data...
Calculating metrics...
Saving metrics to: /jacob.wang/projects/OpenPCDet/output/cfgs/nuscenes_models/cbgs_voxel0075_largekernel_centerpoint/default/eval/epoch_3/val/default/final_result/data
mAP: 0.5352
mATE: 0.3112
mASE: 0.2562
mAOE: 0.3920
mAVE: 0.2666
mAAE: 0.1928
NDS: 0.6257
Eval time: 95.1s
Per-class results:
Object Class AP ATE ASE AOE AVE AAE
car 0.815 0.193 0.165 0.217 0.254 0.198
truck 0.523 0.313 0.181 0.160 0.226 0.217
bus 0.663 0.336 0.191 0.066 0.395 0.266
trailer 0.368 0.569 0.206 0.531 0.203 0.202
construction_vehicle 0.182 0.747 0.429 0.920 0.113 0.284
pedestrian 0.823 0.152 0.279 0.441 0.245 0.111
motorcycle 0.378 0.229 0.246 0.400 0.492 0.254
bicycle 0.338 0.189 0.263 0.663 0.205 0.009
traffic_cone 0.645 0.159 0.327 nan nan nan
barrier 0.616 0.224 0.277 0.130 nan nan
2023-05-09 04:02:49,279 INFO ----------------Nuscene detection_cvpr_2019 results-----------------
***car error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.19, 0.16, 0.22, 0.25, 0.20 | 71.75, 81.95, 85.26, 87.11 | mean AP: 0.8152142922121536
***truck error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.31, 0.18, 0.16, 0.23, 0.22 | 35.86, 52.21, 58.52, 62.50 | mean AP: 0.5226909579459581
***construction_vehicle error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.75, 0.43, 0.92, 0.11, 0.28 | 1.41, 10.28, 25.12, 35.79 | mean AP: 0.1815073390940494
***bus error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.34, 0.19, 0.07, 0.40, 0.27 | 42.85, 65.74, 77.18, 79.46 | mean AP: 0.6630854322478308
***trailer error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.57, 0.21, 0.53, 0.20, 0.20 | 8.50, 30.04, 48.78, 60.01 | mean AP: 0.36832192626705307
***barrier error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.22, 0.28, 0.13, nan, nan | 51.43, 61.96, 65.58, 67.42 | mean AP: 0.6159528156859566
***motorcycle error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.23, 0.25, 0.40, 0.49, 0.25 | 30.57, 38.88, 40.58, 41.37 | mean AP: 0.37848749504321455
***bicycle error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.19, 0.26, 0.66, 0.20, 0.01 | 31.78, 33.94, 34.52, 35.12 | mean AP: 0.3383919265973568
***pedestrian error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.15, 0.28, 0.44, 0.24, 0.11 | 80.02, 81.63, 83.01, 84.66 | mean AP: 0.8233015015580978
***traffic_cone error@trans, scale, orient, vel, attr | AP@0.5, 1.0, 2.0, 4.0
0.16, 0.33, nan, nan, nan | 61.17, 62.76, 64.93, 69.26 | mean AP: 0.6452905854214035
--------------average performance-------------
trans_err: 0.3112
scale_err: 0.2562
orient_err: 0.3920
vel_err: 0.2666
attr_err: 0.1928
mAP: 0.5352
NDS: 0.6257
2023-05-09 04:02:49,280 INFO Result is saved to /jacob.wang/projects/OpenPCDet/output/cfgs/nuscenes_models/cbgs_voxel0075_largekernel_centerpoint/default/eval/epoch_3/val/default
2023-05-09 04:02:49,280 INFO ****************Evaluation done.*****************
Process finished with exit code 0
Please change spatialgroupconv to spatialgroupconvv2
jianingwangind
commented
1 year ago Now it is fixed. Thanks a lot for your detailed explanation:)
Hi, thanks for your released code! I noticed that you released your last work FocalsConv based on both det3d and OpenPCDet, but this time you only released the version on det3d. Do you have any plan to add LargeKernel3D to OpenPCDet? Thanks!