### Semi-supervised Pathological Image Segmentation via Cross Distillation of Multiple Attentions --- - architecture ![image](https://hackmd.io/_uploads/Sy4zyMC66.png) --- #### Tri-branch decoder - Channel attention (CA branch) ![image](https://hackmd.io/_uploads/B1iglMRp6.png) - Spatial attention (SA branch) ![image](https://hackmd.io/_uploads/SJZfeG0ap.png) - Channel and Spatial attention (CSA branch) - CA block followed by an SA block --- #### Cross Decoder Knowledge Distillation - T-softmax (label smoothing? -> Feature) ![image](https://hackmd.io/_uploads/BJctYzRTT.png) --- - Uncertainty minimization (Entropy minimization) ![image](https://hackmd.io/_uploads/rysQOMRp6.png) --- - Ablation study ![image](https://hackmd.io/_uploads/HyvcgmATa.png) --- ### TPRO: Text-Prompting-Based Weakly Supervised Histopathology Tissue Segmentation --- - Framework ![image](https://hackmd.io/_uploads/SkKgvHC6a.png) --- - Pixel-Label correlation ![image](https://hackmd.io/_uploads/Sk6yor06T.png) --- #### Pseudo-label generation - min-max normalization ![image](https://hackmd.io/_uploads/S11TVI06a.png) ![image](https://hackmd.io/_uploads/HJp6V8R6p.png) --- - Pseudo-label Results ![image](https://hackmd.io/_uploads/rkwkbICpT.png) --- - Results ![image](https://hackmd.io/_uploads/H1tZZIC66.png) --- - PistoSeg Results ![image](https://hackmd.io/_uploads/rJJNMI0ap.png) --- ---
{"description":"type: slide","title":"CDMA & TPRO","contributors":"[{\"id\":\"4724c2c9-74e7-418c-87c3-79a619b97461\",\"add\":1472,\"del\":1}]"}
    145 views