CrosSCLR
CrosSCLR copied to clipboard
The Official PyTorch implementation of "3D Human Action Representation Learning via Cross-View Consistency Pursuit" in CVPR 2021
I have run your code and the linear evaluation accuracy is only 45.84%. When I set the "param.requires_grad =True" in line 44,49,54 of linear.py, the linear evalution accuracy of NTU60...
Thank you for sharing. Now I have a few questions. What are the specific versions of CUDA and Cudnn configured by CrossSCLR? Every time I execute "python main.py pretrain_crossclr_3views --config...
I only want to obtain the skeletonclr encoder pretained by moco, how can I use the code?
How to implement the t-sne visualization code
when i try:python main.py pretrain_crossclr_3views --config config/CrosSCLR/crossclr_3views_xview.yaml out:RuntimeError: mat1 dim 1 must match mat2 dim 0 Parameters: {'work_dir': 'work_dir/crossclr_3views/1_xview_frame50_channel16_epoch300_cross150', 'config': 'config/CrosSCLR/crossclr_3views_xview.yaml', 'phase': 'train', 'save_result': False, 'start_epoch': 0, 'num_epoch': 50, 'use_gpu':...
In the forward function of CrosSCLR, the key parameters in joint, bone, motion view are all momentum updated. However, in cross_training function, only key parameters in motion view are momentum...
Thanks for your sharing and your work is very excellent. I'd like to know how to fine-tune CrosSCLR and have the fine-tuning step been included in your code? What is...