search

YudeWang / SEAM

Self-supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation, CVPR 2020 (Oral)
MIT License
539 stars 97 forks source link

Background threshold? #31

Closed whatsups closed 3 years ago

whatsups commented 3 years ago

I notice that you traverse all background threshold options and give the best mIoU of pseudo labels, this setting assumes that the ground truth masks are available during pseudo label generating. However, in practice, if the gt masks are available, why don't we just use these gt labels? So I think a background threshold selection strategy without depending on gt masks is needed here for practice. What do you think of it? Thanks！

YudeWang commented 3 years ago

@pigcv89 Good question, and that's why I am working on semi-supervised semantic segmetnation now. It is inevitable to get some pixel-level annotations to evaluate the trained model. However, the amount of mask is always limited. It is feasible to train a model by image-level labels and select best threshold on pixel-level labels val set. Then generate pixel-level pseudo labels for those image-level labeled images and retrain a model by the combination of pseudo labels and gt mask.

whatsups commented 3 years ago

Thanks for your reply，it's very helpful to me.