Distance Matters in Human-Object Interaction Detection

This repo contains the implementation of our ACM MM 2022 paper titled "Distance Matters in Human-Object Interaction Detection"

Prerequist

We adopted the lightweight library [Pocket][https://github.com/fredzzhang/pocket], where pytorch 1.9.1 with cuda 11.1 are adopted. Please install this library according to the requirements. Then, please prepare the datasets according to the following links:

• HICO-DET: https://github.com/fredzzhang/hicodet
• V-COCO: https://github.com/fredzzhang/vcoco

And then download the two datasets in the respective foloders.

Please also clone the DETR codes into this repository https://github.com/fredzzhang/detr

Training

Our codes run on 4 GPUs in parallel.

To train on the HICO-DET dataset, please run

./scripts/train_hicodet.sh

The evaluation results will be automatically printed after training.

To train on V-COCO, please run

./scripts/train_vcoco.sh

Acknowledgement

Our code is built based on UPT, and DETR. Thanks them for their great work!