runpeidong / pointdistiller Goto Github PK

hello, thank you for your wonderful work.
I was wondering how you handle the channel number difference in calculating the KD-Loss? And what kind of loss was used for KD?

Hello~ your recent work seems very impressive. I'd like to try several experiments with your work.
Can you please release the code as soon as possible? I'm really looking forward to it!

When I try to run CUDA_VISIBLE_DEVICES=0 python tools/train.py configs/pointpillars/hv_pointpillars_secfpn_6x8_160e_kitti-3d-3class_student_4x.py --use-kd ，I encounter the following error message. How can I solve it?
2023-05-22 20:03:39,521 - mmdet - INFO - Set random seed to 0, deterministic: False
/home/k/KD/PointDistiller/mmdetection3d/mmdet3d/models/builder.py:53: UserWarning: train_cfg and test_cfg is deprecated, please specify them in model
'please specify them in model', UserWarning)
Traceback (most recent call last):
File "tools/train.py", line 294, in
main()
File "tools/train.py", line 248, in main
test_cfg=cfg.get('test_cfg'))
File "/home/k/KD/PointDistiller/mmdetection3d/mmdet3d/models/builder.py", line 85, in build_model
return build_detector(cfg, train_cfg=train_cfg, test_cfg=test_cfg)
File "/home/k/KD/PointDistiller/mmdetection3d/mmdet3d/models/builder.py", line 55, in build_detector
'train_cfg specified in both outer field and model field '
AssertionError: train_cfg specified in both outer field and model field

Hi! Congratulations! when will you open source the codes? Looking forward to it!

Hi, thanks for the work. Just wondering what the values in Fig. 3(b) actually denote? (e.g. [0.3, 0.2 0.1, ...], from which 0.3 is taken)

In your preprint paper, I see that you have compared your work with 'SESSD'. And I also try to re-implement the 'SESSD' based on OpenPCDet. But there seem many problems in my re-implement versions, so I could not re-produce its result. Therefore, I am very interested in the SESSD you repo have implemented, and I hope the code can be open-sourced as soon as possible

Hi, I would like to know when the source code will open.