lixiangpengcs / self-supervised-video-hashing Goto Github PK

Hello, I managed to reconstruct the result on FCVID with the help of your code, but I met trouble on YFCC. E.g. According to the paper, the reported MAP@5 for 128-bits on YFCC was about 0.38, but I only got 0.18. My training process for YFCC is like this: I recursively used the 9 training splits to feed the network, and the order of the training splits was randomly chosen for each epoch. For retrieval, I concat the 2 test splits as a whole test matrix. I kept the parameters the same with your default setting. Could you please inform me of your training details on YFCC dataset? Thank you very much.

Hello, I wonder what is the data type of your yfcc_test_feats? The others such as fcv_test_feats are numpy, but when I apply your yfcc_test_feats feature in pytorch, here comes the error "from_numpy expects an np.ndarray but got Dataset". And when I try to change it to numpy by running np.array(yfcc_test_feats), the program jammed.... I don't know how to solve this problem.

我对您的工作很感兴趣，如果可以的话，希望您能提供一下SSHV.py文件。我的邮箱是[email protected]

Hello, I didn't find SSVH.py in your submitted codes. I 've got the datasets, but what should I do to run these codes? Thanks.