This folder contains the code necessary for all text-only adaptations of the vision-and-language models CLIP-BERT, LXMERT and VisualBERT.
- The adaptations that involve finetuning of VL models in a text-only setting (
no-visual-features-finetuned-LXMERT
,no-visual-features-finetuned-Wikipedia
) are described in adaptations/data/runs/finetune/README.md. - The adaptations that only require constant visual feature fillers (
avg-visual-features
,zero-image-visual-features
andzeroed-visual-features
) are handled in adaptations/src/generate_visual_features.ipynb. The notebook generates the constant visual features for the adaptations. - The adaptations that involve finetuning visual features to a text-only setting (
finetuned-LXMERT-visual-features
,finetuned-Wikipedia-visual-features
) are described in adaptations/data/runs/finetune-visual-features/README.md