why the final experiments result on test _set is better than it on val_set?

YudeWang / SEAM

Self-supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation, CVPR 2020 (Oral)

MIT License

539 stars 97 forks source link

why the final experiments result on test _set is better than it on val_set? #28

Open mouzhengdong opened 3 years ago

mouzhengdong commented 3 years ago

Hi,thanks for sharing your job. did u train the segmention network with train_set only(not with train_aug set )and get the 64.5 of miou on val_set? and can you explain why the final experiments result on test _set is better than it on val_set, which made me confused?