search

kchengiva / DecoupleGCN-DropGraph

The implementation for "Decoupling GCN with DropGraph Module for Skeleton-Based Action Recognition" (ECCV2020).
Other
102 stars 19 forks source link

How to do training for a new format of joint #6

Open rd20karim opened 3 years ago

rd20karim commented 3 years ago

Thank's for sharing this code. I was wondering if exist a way to use different format of jointand number of joints in training, what would be the changes that we should made in this project. Thank's for your Answer

saniazahan commented 3 years ago

Hi I am also wondering how did you evaluate your model on NW-UCLA data as it has different number of joints. Could you please share your NW-UCLA preprocessing code and/or data. Did you use NTU pretrained model for it?

saniazahan commented 3 years ago

Hi got my answer from this issue. Thanks a lot for sharing your work.

Thanks for your interest. We update our repo. You can git clone the new repo and run CUDA_VISIBLE_DEVICES="0" python main.py --config ./config/nw-ucla/train_nwucla_*.yaml to train on NW-UCLA dataset, where * is the stream name. Note that the /home/share/NW-UCLA/data/all_sqe/ in ./feeders/feeder_nw.py is your directory for NW-UCLA dataset. The preprocess of NW-UCLA dataset is borrow from Chenyang Si, the author of "An Attention Enhanced Graph Convolutional LSTM Network for Skeleton-Based Action Recognition". The trained model is provided as ./save_models/nw-ucla_joint.pt.

Anirudh257 commented 3 years ago

@rd20karim Changing the graph joints should suffice.