We use GAT network to solve the problem of low accuracy of text retrieved images faced in unsupervised cross-modal hash retrieval. We also provide GCN network. The GCN network is experimentally proven to be slightly worse than the GAT network.
We provide the training models of GAT and GCN, you can compare them by yourself.Our work is based on Pytorch 11.3, using two RTX3090s for training
We provide two datasets, MS COCO we use the one provided by DAEH authors, please excuse us Flickr-25K, NUS-WIDE dataset, please refer to DSAH.
MIRFlickr-25K ！[panbaidu[https://pan.baidu.com/s/1o5jSliFjAezBavyBOiJxew#list/path=%2F] password: 8dub]
NUS-WIDE (top-10 concept) ![panbaidu[https://pan.baidu.com/s/1GFljcAtWDQFDVhgx6Jv_nQ#list/path=%2F],password: ml4y]
DJSRH,JDSH,HNH,DSAH,DAEH