README.md

Data Setup

We provide access to our preprocessed data (including extracted features) and preprocessing scripts to replicate our setup.

Preprocessed Data

• Conceptual Captions
• Flickr30k
• GQA
• MS COCO
• NLVR2
• RefCOCO (UNC)
• RefCOCO+ (UNC)
• RefCOCOg (UMD)
• SNLI-VE
• VQAv2

Preprocessing Steps

We rely on the airsplay/bottom-up-attention Docker image to extract image features from Faster R-CNN. This docker file for bottom-up-attention is available on docker hub and can be downloaded with:

sudo docker pull airsplay/bottom-up-attention

For more details about the Docker image, see the LXMERT repository. Our scripts assume the pretrained Caffe models to be stored under snap/pretrained/.

---

Check out the README files for each data set for detailed preprocessing procedures.