HandM article image features by cnn

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, o