Skip to content

3dperceptionlab/TOAD

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

This repository contains the official implementation for the ``Text-driven Online Action Detection´´ (TOAD) method, published in Integrated Computer-Aided Engineering (arXiv).

Requirements & Setup

We provide a Dockerfile to create an reproducible environment using Docker and NVIDIA Docker, though Conda or Python environments may be used as well. The Docker image is built locally using the provided Dockerfile with:

cd scripts
docker build -t toad .

Feature Extraction

CLIP features (frame and text) are preprocessed and stored in .pt files for faster training. You may use the files in the clip directory to extract the features. The files will automatically download the corresponding weights from OpenAI API.

Annotations

You may find the annotations we used processed in a pickle file in the annotations directory. You may use the PyTorch dataset class in the dataset directory as an example to open these files.

Running

The configuration files in the config folder provide the necessary parameters to run the code. The main.py file is the entry point to run the code. Please use:

bash scripts/run_train.sh <config_file>

Citation

If you find our paper useful, please consider citing our work.

Contact

If you have any questions, please contact me at: (mbenavent@dtic.ua.es)

About

[ICAE 2025] Official Implementation of "Text-driven online action detection"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors