Semi-supervised-learning

Semi-supervised-learning copied to clipboard

While working on the code, I noticed that there are no global loggers set up in the codebase. Currently, loggers are only tied to the trainer, which has led to...

pizhn

作者您好，阅读了你们发表的SoftMatch论文，感觉受益匪浅。同时我也有一些小疑问：①论文中提到的“伪标签置信度分布的可视化”这里相应的代码实现在哪里呢；②如何判断一个伪标签是否正确？伪标签的置信度数值在[0, 1]之间，而真实的标签只能是1或0... 如果能给予任何指导，将不胜感激！

BranStarkkk

# calculate entropy loss if mask.sum() > 0: ent_loss, _ = entropy_loss(mask, logits_x_ulb_s, self.p_model, self.label_hist) else: ent_loss = 0.0 # ent_loss = 0.0 total_loss = sup_loss + self.lambda_u * unsup_loss...

yumudry

Is this a bug of algorithm mean teacher?

1

https://github.com/microsoft/Semi-supervised-learning/blob/6cf697e6968f9f77040a9e0162306436a91a00dc/semilearn/algorithms/meanteacher/meanteacher.py#L54-L56 we get outs_x_ulb_s in line 54, but we use outs_x_ulb_w in line 55 and 56? 

hellojinjie

About ImageNet127 dataset

1

Hello! I've noticed that you have provided config files in config/classic_cv_imb about imagenet127 dataset. However, there's no descriptions about how to get the dataset. Is this dataset the same as...

Sheldon04

Feature request: Meta pseudo labels

1

Great benchmark! I was wondering is there any plan to implement [Meta Pseudo Labels](https://arxiv.org/pdf/2003.10580) in the repo?

pizhn