search

LiheYoung / ST-PlusPlus

[CVPR 2022] ST++: Make Self-training Work Better for Semi-supervised Semantic Segmentation
https://arxiv.org/abs/2106.05095
MIT License
233 stars 33 forks source link

Question about the "Comparison with state-of-the-art methods" section #4

Closed chaochao42 closed 3 years ago

chaochao42 commented 3 years ago

Hello, thank for your insightful and ingenious work. After thorough reading, I have a simple question that why CPS(Semi-Supervised Semantic Segmentation with Cross Pseudo Supervision) is not compared with ST++, since CPS is a previous SOTA work?

LiheYoung commented 3 years ago

Hi, thanks for your question.

Our work is concurrent with CPS. By the time of our submission, CPS has not yet been available publicly. CPS is available on arXiv on June 2, while ours is submitted on May 28 and available on arXiv on June 9.

Actually, our ST++ performance is slightly better than CPS on Pascal VOC across most settings, especially when considering CPS adopts a ResNet with the deep stem block, which is stronger than our classical ResNet. As for Cityscapes, since CPS introduces extra training strategies, such as OHEM and SyncBN, it is hard to be directly compared.

We may add the comparisons in our future version.

chaochao42 commented 3 years ago

Ok, thank you for your answer.