请教大大一个问题：FileNotFoundError: [Errno 2] No such file or directory: './data/ti_80M_selected.pickle'

YyzHarry / imbalanced-semi-self

[NeurIPS 2020] Semi-Supervision (Unlabeled Data) & Self-Supervision Improve Class-Imbalanced / Long-Tailed Learning

https://arxiv.org/abs/2006.07529

MIT License

736 stars 116 forks source link

Closed mechsihao closed 3 years ago

mechsihao commented 3 years ago

作者大大你好，想请教下，我看论文里整个self-training阶段的总loss为：那这个 \omega 是怎么定的呢？还有半监督学习的总轮数epoch的确定方法。

YyzHarry commented 3 years ago

Hi - thank you for your interest!

For ti_80M_selected.pickle, you can find the file here. The link is also provided in the README.
\omega: We simply use 1 for all experiments. It's possible that a better selected \omega can lead to better results. As for the epoch number, it is the same as the one without unlabeled data (you can refer to our code, e.g., train_semi.py).