Semantic Segmentation on PhenoBench (val)
85.55mIoUDAS-SK
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DAS-SKType=CNN, Param. (M)=10.678, GFLOPs=45.00, log(Param.)=1.028, Efficiency (%)=10.09, FPS=10.33, Input image size=1024 x 10242026.02 | 85.55 | 4.67 | — | — | — | |
| DeepLabV3+ ResNet101Type=CNN, Param. (M)=45.670, GFLOPs=233.89, log(Param.)=1.660, Efficiency (%)=1.25, FPS=7.75, Input image size=1024 x 10242026.02 | 85.52 | 4.64 | — | — | — | |
| UNet ResNet34Type=CNN, Param. (M)=24.436, GFLOPs=125.63, log(Param.)=1.388, Efficiency (%)=2.64, FPS=15.19, Input image size=1024 x 10242026.02 | 85.48 | 4.6 | — | — | — | |
| DeepLabV3 ResNet101Type=CNN, Param. (M)=58.626, GFLOPs=964.56, log(Param.)=1.768, Efficiency (%)=0.24, FPS=2.42, Input image size=1024 x 10242026.02 | 84.98 | 4.1 | — | — | — | |
| PSPNet ResNet50Type=CNN, Param. (M)=24.314, GFLOPs=46.87, log(Param.)=1.386, Efficiency (%)=Baseline, FPS=22.05, Input image size=1024 x 10242026.02 | 80.88 | — | — | — | — | |
| WaWBackbone=BioCLIP, Approach=PB (WaW)2026.04 | 59.34 | — | 98.85 | 68.06 | 11.13 | |
| PASTABackbone=CLIP ViT-B-32 LAION2b, Approach=PB (Ours)2026.04 | 52.43 | — | 98.59 | 51.83 | 6.89 | |
| PASTABackbone=CLIP ViT-B-16, Approach=PB (Ours)2026.04 | 50.85 | — | 98.59 | 48.72 | 5.23 | |
| PASTABackbone=BioCLIP, Approach=PB (Ours)2026.04 | 50.09 | — | 98.59 | 44.93 | 6.76 | |
| PASTABackbone=DINOv3 ViT-Small, Approach=PB (Ours)2026.04 | 47.76 | — | 98.59 | 38.71 | 5.97 | |
| PASTABackbone=DINOv3 ConvNeXt-Tiny, Approach=PB (Ours)2026.04 | 47.27 | — | 98.59 | 35.59 | 7.61 |