Pairwise CNN-Transformer Features

This is the official implementation code for Pairwise CNN-Transformer Features for Human-Object Interaction Detection. [paper]

Status

We don't update the code anymore. Please contact [email protected] (emil) with any questions. If you are interested in our work, please read the UPT code first; reproducing our work is straightforward.

Training and Testing

Refer to launch_template.sh for training and testing commands with different options.

To test the PCT model on HICO-DET, you can either use the Python utilities UPT implemented or the Matlab utilities provided by Chao et al.. For V-COCO, we did not implement evaluation utilities, and instead use the utilities provided by Gupta et al.. Refer to these instructions for more details.

Model Zoo

UPT provides weights for fine-tuned DETR models to facilitate reproducibility. To attempt fine-tuning the DETR model yourself, refer to this repository.

Model	Dataset	Default Settings	PCT Weights
PCT-R50	HICO-DET	(33.63, 28.73, 35.10)	weights
PCT-R101	HICO-DET	(33.79, 29.70, 35.00)	weights

Model	Dataset	Scenario 1	Scenario 2	PCT Weights
PCT-R50	V-COCO	59.4	65.0	weights
PCT-R101	V-COCO	61.4	67.1	weights

Acknowledgement

Many thanks to Researcher Zhang for the valuable advice.