about extracting features

Question

about extracting features

jumptoliujj opened this issue 3 years ago · comments

I check out your code(https://github.com/yl-1993/hfsoftmax.git) and switch to branch ltc for extracting feature using pretrained model(https://drive.google.com/file/d/1eKsh7x-RUIHhIJ1R9AlUjsJdsdbh2qim/view) you provide in learn-to-cluster(https://github.com/yl-1993/learn-to-cluster/blob/master/DATASET.md). But I can not extract good features(diff from part0_train.bin you provide) and then the knn generated by my extracting features is not good as well.What's wrong with this situation？

Thanks.

buptzyb · Answer 1 · Wed May 26 2021 17:29:31 GMT+0800 (China Standard Time)

Hi, I have the same question as you. The features extracted by the pretrained model are much worse than those provided. Besides, I found that the provided features (such as part0_train.bin) are normalized, so I think the author may not use this script (https://github.com/yl-1993/hfsoftmax.git) to extract the features. Do you have any solutions?
@jumptoliujj @yl-1993

jumptoliujj · Answer 2 · Sun May 30 2021 16:07:37 GMT+0800 (China Standard Time)

I have no idea. Please tell me the solution if you find out. Thanks.
@dubdynasty

buptzyb · Answer 3 · Tue Jun 01 2021 11:25:13 GMT+0800 (China Standard Time)

No idea, too... Perhaps we have to train the feature extractor on part0 by ourselves.