classicvalues / bgan

BGAN

To run this codes, you should do follow things:

1. extract resnet feature and then run create_S.py to construct similarity matrix.
   You can download cifar-10.mat (here) training data and cifar_KNN.npz from (here)
2. download vgg19 pretrained model on ImageNet based tensorflow here .
3. run this command 'python BGAN.py 32' to train network and then generate 32 bit codes
4. after training done, you can run evalutate.py to calculate MAP. HIT: you need to change some paths in evalutate.py.