search

zhmiao / OpenLongTailRecognition-OLTR

Pytorch implementation for "Large-Scale Long-Tailed Recognition in an Open World" (CVPR 2019 ORAL)
BSD 3-Clause "New" or "Revised" License
839 stars 128 forks source link

One thing about the normalization #26

Closed tonghe90 closed 5 years ago

tonghe90 commented 5 years ago

Thanks for sharing the code. I have one question about the squashing function + cross entropy loss. Do you have some experiments about using softmax + cross entropy loss? Or other normalization method

zhmiao commented 5 years ago

Hello @tonghe90 , thanks for asking. Yes, we have tested pure softmax + cross entropy and simple cosine normalization + cross entropy. Neither of them were as good as squashing because squashing is relatively more relaxed compared to cosine, and normalized than direct feature. We are sorry wer did not report the comparison in the paper due to space limitations.