### Online Easy Example Mining forWeakly-supervised Gland Segmentation from Histology Images (OEEM) --- ### motivation - existing WSSS methods do not perform well in gland segmentation - unique charateristics in glandular datasets - ex. morphological homogeneity and low contrast among different tissues --- ### proposed principal - focuses the network on credible supervision signals rather than noisy signals, thus reducing the influence of false predicitons in pseudo-masks - propose the OEEM to reweight the loss map for better usage of credible and clean supervision --- - OEEM architecture  --- - reweight loss weight  --- - confidence  - The noises woth high confidence on the false category have high loss values, and those pixels supervised by clean labels have lower values  --- - combine normalised loss and max confidence  --- - results  --- - results  --- ---