ROAD Dataset Research

Dockerized Research Repository for Action Classification.

Baseline (3D-RetinaNet)

Getting Started

1. docker compose build
2. docker compose up <service>, where the services include our action classifier, tracker, video builder, post-processing techniques, and parts of our evaluation scheme.
3. docker exec -it <service> /bin/bash to begin developing in the container. Note, code is ran using python main.py --config /path/to/config.yml. Please refer to each module's README for more details.

Repo Structure

Most components are dockerized, we transfer data between each of the components via JSON files and CSVs.

For submodules, please refer to their setup instructions.

This repository is no longer maintained, but it will be kept available for those who are interested in doing future research in Action Classifcation.