Cell Instance Segmentation on PanNuke (3-fold cross-val average)
63.62bPQMulti-Task Image-to-Image Schrödinger Bridge
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Multi-Task Image-to-Image Schrödinger BridgeBackbone=U-Net, Pre-training=unconditional ADM on ImageNet, Diffusion steps (T)=502026.04 | 63.62 | 81.06 | 80.11 | 84.58 | |
| CellViT+procPost-processing=true, Backbone=SAM-H2026.04 | 62.21 | 79.46 | 78.47 | 83.57 | |
| CellViTPost-processing=false, Backbone=SAM-H2026.04 | 61.49 | 80.37 | 83.28 | 79.84 | |
| U-Net+RSupervision type=instance masks + reverse distance maps2026.04 | 58.37 | 79.42 | 79.72 | 79.08 | |
| U-NetSupervision type=binary instance masks2026.04 | 57.99 | 79.3 | 79.89 | 78.74 |