multimodal. A collection of multimodal datasets, and visual features for VQA and captionning in pytorch. Just run "pip install multimodal"