iit-pavis / uhar_skeletal_laplacian

Code for the BMVC'21 Oral paper "Unsupervised Human Action Recognition with Skeletal Graph Laplacian and Self-Supervised Viewpoints Invariance"

Unsupervised Human Action Recognition with Skeletal Graph Laplacian and Self-Supervised Viewpoints Invariance

This repository provides the Pytorch code for our work accepted to BMVC 2021 as an Oral Presentation.

Code repository with training script for the NTU-60 and NTU-120 datasets.

Requirements

• Numpy
• Scikit-learn
• Pytorch
• Tqdm
• Wandb

The code has been run on PyTorch version 1.6.0, and we therefore recommend this version.

For any questions, feel free to contact [email protected]

Dataset preprocessing

• NTU-60

  • Download zip file containing raw skeleton data here
  • Extract the nturgb+d_skeletons folder, contained inside zip file, to "dataset/raw/ntu_60"
  • Execute NTU60_dataset_preprocessing.py

• NTU-120

  • Download zip files containing raw skeleton data here and here
  • Extract the nturgb+d_skeletons folders, contained inside zip files, to "dataset/raw/ntu_120"
  • Execute NTU120_dataset_preprocessing.py

Citation

@inproceedings{UHAR_BMVC2021,
   title={{Unsupervised Human Action Recognition with Skeletal Graph Laplacian and Self-Supervised Viewpoints Invariance}},
   author={Paoletti, Giancarlo and Cavazza, Jacopo and Beyan, Cigdem and Del Bue, Alessio},
   booktitle={The 32nd British Machine Vision Conference (BMVC)},
   year={2021},
}

Disclaimer

The software is provided "as is", without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose and noninfringement. In no event shall the authors, PAVIS or IIT be liable for any claim, damages or other liability, whether in an action of contract, tort or otherwise, arising from, out of or in connection with the software or the use or other dealings in the software.

LICENSE

This project is licensed under the terms of the MIT license.

uhar_skeletal_laplacian's Issues

Linear_evaluation_protocol.py is required

About the evaluation setup

Hi authors.

Thank you for this outstanding project. It has helped me a lot with this kind of ML problem. I am having a problem of understanding and reproducing the result. Since the implementation of Linear Evaluation Protocol has not been provided and has not been described in the paper, could you please tell me more about it? Is that fine-tuning the complete model (including the backbone) or only fine-tuning fully connected layers?

Thank you authors,