Zxl19990529 / Class-driven-Scribble-Promotion-Network

The official implementation of AAAI2024 paper of "Scribble Hides Class: Promoting Scribble-based Semantic Segmentation with its Class Label"
12 stars 1 forks source link

How to generate pseudo labels? #2

Open Fluestech opened 1 month ago

Fluestech commented 1 month ago

Hi,

Great job on the project!

Could you let me know where I can find the BMP code?

Thanks!

Zxl19990529 commented 1 month ago

Hi, I‘m sorry that I forgot to release the pseudo labels used in the paper, here is the link google drive . I also updated the Readme of this project. However, I do not have the authority to publicly disclose the code of BMP. It's recommanded to ask for the code from the first author of "Branches Mutual Promotion for End-to-End Weakly Supervised Semantic Segmentation".

Fluestech commented 1 month ago

Hi @Zxl19990529 ,

Many thanks for this!

Btw, shouldn’t the results of weakly SSS generally be lower than those of fully SSS? I'm curious about how much lower they will be.

Thanks

Zxl19990529 commented 1 month ago

Yes, the results of weakly SSS are lower than fully supervised ones. In general, the rank is : Image supervised< scribble supervised< Fully supervised I have uploaded the quantization results of the fully supervised ones in Readme page for reference.

Fluestech commented 1 month ago

OK, thanks!