This is the implementation of Image2Tripplets, including a forward and backward model.
conda install --yes --file requirement.yaml
- Prepare the dataset. You can use gui.py in the hoi_dataset folder to annotate the HOI, but you should first annotate the objects and humans in the images in the format of the VOC dataset.
- Prepare word embeddings. You can follow the instruction in the readme.md under the word_embeddings folder.
Backward process
python backward_model.py
Forward process
python forward_model.py
python demo.py