SCAN Faster R-CNN Image Features

This dataset contains precomputed MS-COCO and Flickr30K Faster R-CNN image features, which are all the data needed for reproducing the experiments in "Stacked Cross Attention for Image-Text Matching", our ECCV 2018 paper. We use splits produced by Andrej Karpathy. The raw images can be downloaded from their original sources http://nlp.cs.illinois.edu/HockenmaierGroup/Framing_Image_Description/KC..., http://shannon.cs.illinois.edu/DenotationGraph/ and http://mscoco.org/.

The precomputed image features of MS-COCO are originally from https://github.com/peteanderson80/bottom-up-attention. The precomputed image features of Flickr30K are extracted from the raw Flickr30K images using the bottom-up attention model from https://github.com/peteanderson80/bottom-up-attention.