Pre Trained Models

Pre Trained Models

This folder contains pre trained models weights for the different models trained and mentioned in the paper.

There 3 sub folder within this folder, namely:

1. NTU 120 (Contains model weights for NTU 120 datasets)
2. skeletics-152 (Contains models weights for skeletics-152 dataset)
3. skeleton-mimetics (Contains models weights for skeleton-mimetics dataset)

The weights are stored in standard .pt format, and can be easily used using the method mentioned in respective model github repo.

NOTE: Pre trained model weights for VaCNN on NTU 120 can be dowloaded from here. Cross Subject and Cross Setup

Naming Convention:

The weight files are named in according to the dataset on which it is trained, stream name and epoch number.

For example, ntu_xset_2sSDGCN_bone_epoch_31.pt file contains weight for 2s-SDGCN model's bone stream trained on NTU-120 Cross setup(Xset) till epoch 31

And similarly all other model weight files are named

NOTE: MsG3D requires 2 separate files for weights and checkpoints, which are added using " weights " and " checkpoint " in the file name

Github repo link for Models:

Following is the list of different models codes. Instruction to use pretrained weights are given in the README of each of the models:

• MsG3D
• ShiftGCN
• VaCNN
• DgNN
• GCN-NAS