HT-ConvNet

This is the official repository for Hierarchical Temporal Convolution Network: Towards Privacy-Centric Activity Recognition, our paper published at the International Conference on Ubiquitous Computing and Ambient Intelligence (UCAmI)

Prerequisites

Python >= 3.6

PyTorch >= 1.1.0

PyYAML, tqdm `

Data Preparation

Download datasets.

Download the JHMDB and SHREC datasets using the links below:

JHMDB raw data download link:   http://jhmdb.is.tue.mpg.de/challenge/JHMDB/datasets
SHREC raw data download link:   http://www-rech.telecom-lille.fr/shrec2017-hand/

Process datasets.

Use the preprocessing code in the data processing folder to process the data and put them in the data folder.

Or you can get the already processed data directly from this GitHub repo

A sample of the processed file is currently in the data folder, please replace it.

Training

For JHMDB, run python train.py --batch-size 512 --epochs 600 --dataset 0 --lr 0.001 | tee train.log

For SHREC coarse, run python train.py --batch-size 512 --epochs 600 --dataset 1 --lr 0.001 | tee train.log

For SHREC fine, run python train.py --batch-size 512 --epochs 600 dataset 2 --lr 0.001 | tee train.log

Testing

To test the trained model, bring the saved model to the main directory and pass its name as an arg for the model-path or simply pass the path to where the model was saved

For JHMDB, run python test.py --model-path model.pt --dataset 0

For SHREC coarse, run python test.py --model-path model.pt --dataset 1

For SHREC fine, run python test.py --model-path model.pt --dataset 2

To force the model to be loaded with CPU run python test.py --model-path model.pt --dataset 0 --no-cuda

Action Recognition in Real-time with HT-ConvNet