Public code for https://github.com/judithfan/visual_communication_in_context.
Download dependencies:
pip install -r ./requirements.txt
Download data:
Go to https://figshare.com/projects/Visual_Communication_in_Context/38321. Download all zip files (should be 3) to ./visual_module/data folder. Unzip all 3 files.
To train using all 5 splits, run:
python run_cv5.py high --out-dir ./trained_models
The first positional argument can be high, mid, or early depending on your chosen adaptor type.
To get similarity metrics and prepare for Bayesian data analysis, run:
python eval_cv5.py ./trained_models --out-dir ./similarity_dump