Semantic Segmentation on MSRS
86.7mIoUTDFusion
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TDFusionwith FusionRegister=true2026.03 | 86.7 | — | — | — | — | — | — | — | — | — | — | 92.9 | — | — | |
| MMDRwith FusionRegister=true2026.03 | 86.5 | — | — | — | — | — | — | — | — | — | — | 93 | — | — | |
| S4Fusionwith FusionRegister=true2026.03 | 85.9 | — | — | — | — | — | — | — | — | — | — | 92.5 | — | — | |
| TDFusion2026.03 | 85.2 | — | — | — | — | — | — | — | — | — | — | 92.1 | — | — | |
| FreqGANwith FusionRegister=true2026.03 | 84.1 | — | — | — | — | — | — | — | — | — | — | 91.4 | — | — | |
| HCLFusewith FusionRegister=true2026.03 | 83.7 | — | — | — | — | — | — | — | — | — | — | 91.3 | — | — | |
| MMDR2026.03 | 83.6 | — | — | — | — | — | — | — | — | — | — | 91.3 | — | — | |
| FreqGAN2026.03 | 80.7 | — | — | — | — | — | — | — | — | — | — | 89.5 | — | — | |
| TUNI-BBackbone=TUNI-B, Params(M)=29.57, FLOPs(G)=42.64, FPS (RTX 4090)=60, FPS (Jetson Orin NX - CUDA)=6, FPS (Jetson Orin NX - TensorRT BF16)=14, Image Resolution=640 x 4802025.09 | 80.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| S4Fusion2026.03 | 79.9 | — | — | — | — | — | — | — | — | — | — | 89.2 | — | — | |
| TUNI-SBackbone=TUNI-S, Params(M)=10.63, FLOPs(G)=17.16, FPS (RTX 4090)=120, FPS (Jetson Orin NX - CUDA)=11, FPS (Jetson Orin NX - TensorRT BF16)=27, Image Resolution=640 x 4802025.09 | 79.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HCLFuse2026.03 | 79.2 | — | — | — | — | — | — | — | — | — | — | 88.7 | — | — | |
| MCMAEPre-train Dataset=IN1K(RGB), Framework=UperNet2025.09 | 79 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UNIV(LoRA)Pre-train Dataset=IN1K+MVIP(RGB+IR), Framework=UperNet2025.09 | 79 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SigmaBackbone=Vmamba-T, Params(M)=48.3, FLOPs(G)=89.5, FPS (RTX 4090)=19, Image Resolution=640 x 4802025.09 | 78.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SemLA2026.03 | 78.7 | — | — | — | — | — | — | — | — | — | — | 88.4 | — | — | |
| IMF2026.03 | 78.7 | — | — | — | — | — | — | — | — | — | — | 88.4 | — | — | |
| TUNI-TBackbone=TUNI-T, Params(M)=4.99, FLOPs(G)=10.82, FPS (RTX 4090)=135, FPS (Jetson Orin NX - CUDA)=16, FPS (Jetson Orin NX - TensorRT BF16)=35, Image Resolution=640 x 4802025.09 | 78.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFormerBackbone=DFormer-B, Params(M)=29.5, FLOPs(G)=41.9, FPS (RTX 4090)=69, FPS (Jetson Orin NX - CUDA)=5, FPS (Jetson Orin NX - TensorRT BF16)=15, Image Resolution=640 x 4802025.09 | 77.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CM-SSMBackbone=EfficientVit-B1, Params(M)=10.34, FLOPs(G)=12.59, FPS (RTX 4090)=114, FPS (Jetson Orin NX - CUDA)=9, FPS (Jetson Orin NX - TensorRT BF16)=42, Image Resolution=640 x 4802025.09 | 77.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TUNI†Backbone=TUNI-S, Params(M)=10.63, FLOPs(G)=17.16, FPS (RTX 4090)=120, FPS (Jetson Orin NX - CUDA)=11, FPS (Jetson Orin NX - TensorRT BF16)=27, Image Resolution=640 x 4802025.09 | 77.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLNet-TBackbone=Mit-B4, Params(M)=130.84, FLOPs(G)=217.85, FPS (RTX 4090)=28, FPS (Jetson Orin NX - CUDA)=1, Image Resolution=640 x 4802025.09 | 76.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SGFNetBackbone=ResNet50, Params(M)=125.25, FLOPs(G)=144.83, FPS (RTX 4090)=42, FPS (Jetson Orin NX - CUDA)=2, Image Resolution=640 x 4802025.09 | 75.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CMXBackbone=Mit-B2, Params(M)=66.57, FLOPs(G)=67.20, FPS (RTX 4090)=63, FPS (Jetson Orin NX - CUDA)=3, FPS (Jetson Orin NX - TensorRT BF16)=13, Image Resolution=640 x 4802025.09 | 75.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InfMAEPre-train Dataset=Inf30(IR), Framework=UperNet2025.09 | 75.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TDFusionBackbone=SegFormer, Training Epochs=3002024.12 | 75.09 | 86.04 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DuGI-MAEBackbone=ViT-B, Model=UperNet2025.12 | 75 | 83.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CMNextBackbone=Mit-B2, Params(M)=58.68, FLOPs(G)=68.70, FPS (RTX 4090)=67, FPS (Jetson Orin NX - CUDA)=3, FPS (Jetson Orin NX - TensorRT BF16)=11, Image Resolution=640 x 4802025.09 | 74.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CFI2026.03 | 74.7 | — | 98.3 | 88.3 | 70.7 | 68.6 | 58.4 | — | — | — | — | — | 64.2 | — | |
| MiLNetBackbone=Mit-B3, Params(M)=92.29, FLOPs(G)=136.32, FPS (RTX 4090)=25, FPS (Jetson Orin NX - CUDA)=2, Image Resolution=640 x 4802025.09 | 74.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MRFSBackbone=SegFormer, Training Epochs=3002024.12 | 74.5 | 84.76 | — | — | — | — | — | — | — | — | — | — | — | — | |
| InfMAEBackbone=ViT-B, Model=UperNet2025.12 | 74.5 | 82.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| EMMABackbone=SegFormer, Training Epochs=3002024.12 | 74.48 | 85.99 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DCINNBackbone=SegFormer, Training Epochs=3002024.12 | 74.35 | 84.11 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SegMIFBackbone=SegFormer, Training Epochs=3002024.12 | 74.25 | 85.73 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAEPre-train Dataset=IN1K(RGB), Framework=UperNet2025.09 | 74.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFormerV2Backbone=DFormerV2-S, Params(M)=26.70, FLOPs(G)=33.90, FPS (RTX 4090)=33, FPS (Jetson Orin NX - CUDA)=2, FPS (Jetson Orin NX - TensorRT BF16)=11, Image Resolution=640 x 4802025.09 | 74.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MURFBackbone=SegFormer, Training Epochs=3002024.12 | 74.08 | 85.03 | — | — | — | — | — | — | — | — | — | — | — | — | |
| A2RNet2026.03 | 74 | — | 98.2 | 88.1 | 68.7 | 68.8 | 55.6 | — | — | — | — | — | 64.4 | — | |
| GMNetBackbone=ResNet101, Params(M)=191.24, FLOPs(G)=195.37, FPS (RTX 4090)=41, FPS (Jetson Orin NX - CUDA)=4, Image Resolution=640 x 4802025.09 | 73.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MCNet-TBackbone=ConvNextV2-B, Params(M)=199.79, FLOPs(G)=278.85, FPS (RTX 4090)=32, Image Resolution=640 x 4802025.09 | 73.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisibleBackbone=SegFormer, Training Epochs=3002024.12 | 73.76 | 83.44 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Conti2026.03 | 73.6 | — | 98.2 | 87.8 | 69.9 | 67.1 | 54 | — | — | — | — | — | 64.6 | — | |
| LUT-Fuse2026.03 | 73.6 | — | 98.2 | 87.8 | 68.7 | 68.1 | 56.4 | — | — | — | — | — | 62.3 | — | |
| IGNet2026.03 | 73.6 | — | 98.2 | 87.2 | 68.5 | 68.8 | 54.1 | — | — | — | — | — | 64.8 | — | |
| TIMFusionBackbone=SegFormer, Training Epochs=3002024.12 | 73.58 | 83.67 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LRRNet2026.03 | 73.4 | — | 98.2 | 88 | 67.6 | 67.9 | 54.8 | — | — | — | — | — | 64.2 | — | |
| MCMAEBackbone=ViT-B, Model=UperNet2025.12 | 73.2 | 81.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DuGI-MAEBackbone=ViT-B, Model=FCN2025.12 | 73.1 | 80.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DCEvo2026.03 | 73.1 | — | 98.2 | 87.8 | 68.7 | 67.9 | 52.4 | — | — | — | — | — | 63.8 | — | |
| SAGE2026.03 | 73 | — | 98.2 | 87.1 | 67.8 | 68 | 54.8 | — | — | — | — | — | 62.1 | — | |
| InfMAEBackbone=ViT-B, Model=FCN2025.12 | 72.6 | 80.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MUFusion2026.03 | 72.2 | — | 98.1 | 86.6 | 66.7 | 67.9 | 50.7 | — | — | — | — | — | 63.4 | — | |
| TarDALBackbone=SegFormer, Training Epochs=3002024.12 | 71.35 | 81.93 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAEBackbone=ViT-B, Model=UperNet2025.12 | 71.3 | 78 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MCMAEBackbone=ViT-B, Model=FCN2025.12 | 70.8 | 79.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ISFMSegmentation Backbone=DeepLabV3+2026.02 | 70.53 | — | 98.15 | 88.66 | 66.3 | 68.69 | 49.88 | 69 | 66.22 | 54.29 | 73.6 | — | — | — | |
| C2RF2026.03 | 70.5 | — | — | — | — | — | — | — | — | — | — | 81.5 | — | — | |
| TEDFusion2026.06 | 69.64 | — | 98.07 | 75.58 | 42.08 | 62.85 | — | — | — | — | — | — | — | — | |
| InfraredBackbone=SegFormer, Training Epochs=3002024.12 | 69.49 | 83.23 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MMDRFuseSegmentation Backbone=DeepLabV3+2026.02 | 69.02 | — | 98.05 | 87.91 | 65.68 | 67.2 | 46.12 | 68.01 | 65.05 | 53.66 | 69.51 | — | — | — | |
| GIFuse2026.06 | 68.9 | — | 98.02 | 75.54 | 40.73 | 61.29 | — | — | — | — | — | — | — | — | |
| Text-IF2026.06 | 68.25 | — | 98.02 | 76.03 | 40.43 | 58.53 | — | — | — | — | — | — | — | — | |
| FusionMambaSegmentation Backbone=DeepLabV3+2026.02 | 67.32 | — | 98.02 | 87.93 | 65.88 | 66.17 | 43.58 | 65.45 | 58.01 | 52.75 | 68.13 | — | — | — | |
| DDRNet2025.12 | 67.3 | 73.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| TextFusion2026.06 | 67.11 | — | 97.94 | 74.35 | 32.69 | 63.48 | — | — | — | — | — | — | — | — | |
| GIFNet2026.06 | 67.08 | — | 97.85 | 73.19 | 35.98 | 61.3 | — | — | — | — | — | — | — | — | |
| DNLNetBackbone=Resnet1012025.12 | 67 | 75.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SemLA2026.06 | 66.16 | — | 97.84 | 73.87 | 32.96 | 59.95 | — | — | — | — | — | — | — | — | |
| DCINN2026.06 | 65.94 | — | 97.78 | 72.42 | 35.98 | 57.6 | — | — | — | — | — | — | — | — | |
| Mask-DiFuserFPS=*, Latency Tier=Tier 1 – Very High Latency, Frozen Downstream Model=SegFormer-B22026.05 | 65.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| U2Fusion2026.06 | 65.79 | — | 97.7 | 70.47 | 34.44 | 60.56 | — | — | — | — | — | — | — | — | |
| VDMUFusion2026.06 | 65.73 | — | 97.8 | 73.1 | 32.92 | 59.07 | — | — | — | — | — | — | — | — | |
| UperNetBackbone=Resnet502025.12 | 65.6 | 74.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DenseFuse2026.06 | 65.45 | — | 97.82 | 73.91 | 24.7 | 65.37 | — | — | — | — | — | — | — | — | |
| FusionProxyFPS=32.16, Latency Tier=Tier 4 – Real-time, Frozen Downstream Model=SegFormer-B22026.05 | 65.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IRFSSegmentation Backbone=DeepLabV3+2026.02 | 65.37 | — | 97.69 | 85.41 | 59.43 | 67.39 | 42.45 | 53.71 | 63.82 | 52.68 | 65.78 | — | — | — | |
| DeeplabV3+Backbone=Resnet502025.12 | 65.2 | 73.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ControlFusionFPS=0.90, Latency Tier=Tier 2 – High Latency, Frozen Downstream Model=SegFormer-B22026.05 | 65.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FISCNetSegmentation Backbone=DeepLabV3+2026.02 | 64.93 | — | 97.67 | 84.29 | 59.89 | 64.55 | 41.8 | 56.98 | 66.17 | 49.05 | 63.97 | — | — | — | |
| DeFusion2026.06 | 64.64 | — | 97.85 | 74.18 | 21.84 | 64.7 | — | — | — | — | — | — | — | — | |
| IVFWSR2026.03 | 64.6 | — | — | — | — | — | — | — | — | — | — | 79.4 | — | — | |
| CDDFuseFPS=1.14, Latency Tier=Tier 2 – High Latency, Frozen Downstream Model=SegFormer-B22026.05 | 64.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DDFMSegmentation Backbone=DeepLabV3+2026.02 | 64.29 | — | 97.64 | 84.84 | 60.31 | 65.09 | 41.37 | 52 | 62.52 | 51.13 | 63.67 | — | — | — | |
| TIMFusion2026.03 | 63.9 | — | 97.5 | 82.6 | 59.7 | 59.7 | 38.9 | — | — | — | — | — | 45 | — | |
| Text-IFFPS=0.85, Latency Tier=Tier 2 – High Latency, Frozen Downstream Model=SegFormer-B22026.05 | 63.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DATFuseSegmentation Backbone=DeepLabV3+2026.02 | 63.63 | — | 97.64 | 84.07 | 60.51 | 63.62 | 40.15 | 56.21 | 61.05 | 48.22 | 61.18 | — | — | — | |
| DeFusionSegmentation Backbone=DeepLabV3+2026.02 | 63.48 | — | 97.79 | 86.21 | 64.91 | 63.9 | 28.33 | 64.58 | 58.39 | 46.51 | 60.67 | — | — | — | |
| UMF-CMGRSegmentation Backbone=DeepLabV3+2026.02 | 63.48 | — | 97.79 | 86.21 | 64.91 | 63.9 | 28.33 | 64.58 | 58.39 | 46.51 | 60.67 | — | — | — | |
| MAEBackbone=ViT-B, Model=FCN2025.12 | 63.4 | 70.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FILMFPS=2.13, Latency Tier=Tier 2 – High Latency, Frozen Downstream Model=SegFormer-B22026.05 | 62.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DDFMFPS=*, Latency Tier=Tier 1 – Very High Latency, Frozen Downstream Model=SegFormer-B22026.05 | 62.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ReCoNet2026.06 | 62.24 | — | 97.52 | 70.89 | 25.94 | 54.6 | — | — | — | — | — | — | — | — | |
| LRRNetSegmentation Backbone=DeepLabV3+2026.02 | 62.14 | — | 97.51 | 82.65 | 63.16 | 65.22 | 36.46 | 52.53 | 52.71 | 49.48 | 59.57 | — | — | — | |
| U2FusionSegmentation Backbone=DeepLabV3+2026.02 | 61.82 | — | 97.84 | 86.2 | 65.51 | 65.74 | 33.68 | 61.29 | 35.29 | 49.49 | 61.33 | — | — | — | |
| SFCFusionSegmentation Backbone=DeepLabV3+2026.02 | 61.82 | — | 97.84 | 86.2 | 65.51 | 65.74 | 33.68 | 61.29 | 35.29 | 49.49 | 61.33 | — | — | — | |
| U2FusionFPS=2.50, Latency Tier=Tier 2 – High Latency, Frozen Downstream Model=SegFormer-B22026.05 | 61.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| From scratchBackbone=ViT-B, Model=UperNet2025.12 | 61.5 | 61.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RPFNetSegmentation Backbone=DeepLabV3+2026.02 | 60.49 | — | 97.15 | 82.23 | 28.43 | 64.93 | 36.06 | 55.41 | 66.86 | 49.19 | 64.18 | — | — | — | |
| TarDALSegmentation Backbone=DeepLabV3+2026.02 | 60.17 | — | 97.67 | 84.7 | 63.1 | 61.96 | 28.41 | 60.56 | 38.98 | 48.56 | 57.62 | — | — | — |