Semantic Segmentation on ADE20K (mIoUSS, mIoUMS)
50.1mIoU (SS)VNCT-S
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| VNCT-S#Param.=44M, FLOPs=1015G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 50.1 | — | — | 51.2 | |
| VMamba-S#Param.=76M, FLOPs=1081G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 49.5 | — | — | 50.5 | |
| VSSD-S#Param.=69M, FLOPs=1002G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 49.2 | — | — | 50.1 | |
| VNCT-T#Param.=54M, FLOPs=950G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 48.8 | — | — | 49.7 | |
| LocalVMamba-T#Param.=57M, FLOPs=970G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 47.9 | — | — | 49.1 | |
| VSSD-T#Param.=53M, FLOPs=941G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 47.9 | — | — | 48.7 | |
| MSVMamba-T#Param.=65M, FLOPs=942G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 47.6 | — | — | 48.5 | |
| Swin-S#Param.=81M, FLOPs=1038G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 47.6 | — | — | 49.5 | |
| VMamba-T#Param.=55M, FLOPs=964G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 47.3 | — | — | 48.3 | |
| VNCT-M#Param.=43M, FLOPs=900G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 46.5 | — | — | 47 | |
| EffVMamba-B#Param.=65M, FLOPs=930G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 46.5 | — | — | 47.3 | |
| ConvNeXt-T#Param.=60M, FLOPs=939G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 46 | — | — | 46.7 | |
| VSSD-M#Param.=42M, FLOPs=893G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 45.6 | — | — | 46 | |
| MSVMamba-M#Param.=42M, FLOPs=875G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 45.1 | — | — | 45.4 | |
| Swin-T#Param.=60M, FLOPs=945G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 44.4 | — | — | 45.8 | |
| EffVMamba-S#Param.=29M, FLOPs=505G, Framework=UPerNet, Pre-training=ImageNet-1K, Training Iterations=160K2026.07 | 41.5 | — | — | 42.1 | |
| DINOv2Pre-training type=self-supervised2026.07 | — | 47.7 | 53.1 | — | |
| DiTFPre-training type=DiTs-based2026.07 | — | 53.6 | 54.8 | — | |
| MREPPre-training type=DiTs-based2026.07 | — | 54.9 | 56.1 | — | |
| TADPPre-training type=SD-based2026.07 | — | 54.8 | 55.9 | — | |
| VPDPre-training type=SD-based2026.07 | — | 53.7 | 54.6 | — | |
| VPD(LS)Pre-training type=SD-based2026.07 | — | 53.7 | 54.4 | — | |
| VPD(R)Pre-training type=SD-based2026.07 | — | 53.1 | 54.2 | — | |
| w/o MREPPre-training type=DiTs-based2026.07 | — | 48.8 | 51.2 | — |