In this work, we focus on skeleton-based one-shot action recognition with diverse occlusion scenarios. In order to get the realistic synthesized occluded dataset for NTU-120, NTU-60 and Toyota Smart Home, please send the license agreement of the original datasets to pengkunyu1013@gmail.com. After checking the datasets will be sent via email. Thank you for your interests!

The required packages are listed in the environment.yml, please create conda env accordingly.

The dataloaders can be found in the dataloader folder, before the running of the code please change the corresponding path accordingly in the dataloader.

First, execute, export DATASET_FOLDER="$(pwd)/data" Then, python train_original.py If you are using Trans4SOAR model please make sure you use one gpu for training.

https://drive.google.com/drive/folders/1E6O9S5Z-ZWTg6iHqZnTM-o-oX9Ica9Bk?usp=share_link

In uploading

In uploading