Some issues regarding the dataset

thuml / HashNet

Code release for "HashNet: Deep Learning to Hash by Continuation" (ICCV 2017)

MIT License

241 stars 84 forks source link

Some issues regarding the dataset #20

Closed yuewuqing2224 closed 5 years ago

yuewuqing2224 commented 6 years ago

Hi, I just looked into the pytorch version of your code and I wonder if you could help clarify some points.

(1) ImageNet-100 train.txt consists of 13000 images. So I assume the reported results are from using 130 images per class during training?

(2) NUS_WIDE in caffe has 81 classes but only 21 classes in pytorch folder. Are the files in pytorch folder the result of selecting only the images with the 21 most frequent labels?

Thank you for answering my questions.

caozhangjie commented 6 years ago

(1) Yes, you are right. (2) I'm sorry we put the wrong dataset in NUS_WIDE directory. We will change them. Thank you.