(This repository is under development)

Code for various methods implemented for HOI detection :

1. (tensorflow) Using only pretrained (on charades) I3D features visual features only.
2. (tensorflow) Using CAD120 precomputed features (Input of GPNN model) only.
3. (tensorflow) Using combined visual-pose features by fusing I3D features with pose heatmaps.
4. (tensorflow) Using GT human pose and bounding box coordinates
5. (pytorch) End to End model with I3D and GT human pose and bounding box coordinates