search

chenqi1126 / SIPE

[CVPR 2022] Self-supervised Image-specific Prototype Exploration for Weakly Supervised Semantic Segmentation
MIT License
73 stars 10 forks source link

Is your final result after CRF #6

Closed Andylijinpan closed 1 year ago

Andylijinpan commented 1 year ago

Because the result I reproduced (after irnet processing), the Miou of train is 68. X, and the result of cam is consistent with your 58.5

chenqi1126 commented 1 year ago

Hi @Andylijinpan,

The pseudo labels on train set do not use extra CRF after irn. Your result (68.X) looks like no problem.

Andylijinpan commented 1 year ago

@chenqi1126 谢谢你的回答，大佬。 对于irnet后的训练集结果68.x，我担心我这个结果通过deeplabv2训练完可能达不到和你的结果一样高，我的这个结果真的可以吗？

Andylijinpan commented 1 year ago

@chenqi1126 另外，大佬方便给一下后半段v2训练的代码吗，我好直接用你的权重

chenqi1126 commented 1 year ago

利用伪标签训练全监督模型的代码是直接使用现有项目，你可以在readme里面找到，只需要修改路径就可以运行。

Andylijinpan commented 1 year ago

利用伪标签训练全监督模型的代码是直接使用现有项目，你可以在readme里面找到，只需要修改路径就可以运行。 谢谢您的回答

henry227 commented 9 months ago

Hello, I want to use the pseudo-label after irnet training to train the fully supervised model of v2, is it the path to replace the real dataset with pseudo-label