search

WangGodder / deep-cross-modal-hashing

Deep learning cross modal hashing in PyTorch
MIT License
102 stars 25 forks source link

CMHH does NOT match the paper #4

Open iTomxy opened 3 years ago

iTomxy commented 3 years ago

Hi, I have run you codes of CMHH on flickr of 64-bit hash, with data & CNN-F weights provided by you, and it's stuck at:

where mAP (i->t) is 0.024 lower than the paper, which is not a small degrade. Also, I draw the histogram of Hamming distances on similar pairs like Fig.6 in the paper, and it looks more like the DCMH's one in the paper rather than the CMHH's one.

So how can the CMHH result be re-produced ?

WangGodder commented 3 years ago

the paper of CMHH did not tell the hyper-parameter of network, and author did not open source code. all hyper-parameters in current code is not the correct parameters.