Open chullhwan-song opened 5 years ago
imagenet에 클래스마다 이미지 수
● We know that we should use at least 500k images and at least 127 classes ● It will probably work well to skip unrelated classes. ● We also know that labeled pretraining seems to outperform other methods.
https://arxiv.org/abs/1608.08614 https://github.com/minyoungg/wmigftl