Stereo Matching on Middlebury
1.1Bad Pixel Rate (Thresh 2.0)FoundationStereo
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| FoundationStereoTraining=strong multi-dataset, Reference Type=paper*2026.06 | 1.1 | — | 1.67 | — | — | — | — | |
| StereoFactoryBackbone=FoundationStereo2026.06 | 2.29 | — | 2.19 | — | — | — | — | |
| FoundationStereoBackbone=FoundationStereo, Training=strong multi-dataset2026.06 | 2.69 | — | 2.43 | — | — | — | — | |
| DEFOM-Stereo2025.01 | 5.02 | — | 1.27 | — | — | 7.73 | — | |
| StereoFactoryBackbone=NMRF-SwinT2026.06 | 5.26 | — | 3.3 | — | — | — | — | |
| FoundationStereoTraining=SceneFlow2026.06 | 5.5 | — | 3.85 | — | — | — | — | |
| FoundationStereoBackbone=FoundationStereo2026.06 | 5.5 | — | 3.85 | — | — | — | — | |
| Selective-IGEV2025.01 | 6.04 | — | 1.54 | — | — | 9.26 | — | |
| MTDzero-shot=true2026.05 | 6.1 | — | — | — | — | — | — | |
| DLNR2025.01 | 6.98 | — | 1.91 | — | — | 10.2 | — | |
| NMRF-SwinTTraining=strong multi-dataset2026.06 | 7.03 | — | 4.06 | — | — | — | — | |
| IGEV++2025.01 | 7.19 | — | 1.83 | — | — | 10.2 | — | |
| IGEVStereo2026.04 | 7.2 | — | — | — | — | — | — | |
| NMRF (Ours)Zero-shot generalization=true, Trained on=SceneFlow2024.03 | 7.5 | — | — | — | — | — | — | |
| NMRFzero-shot=true2026.05 | 7.5 | — | — | — | — | — | — | |
| NMRF2026.06 | 7.5 | — | 5.15 | — | — | — | — | |
| Selective-IGEVTraining=strong multi-dataset2026.06 | 7.5 | — | 4.65 | — | — | — | — | |
| DKT-RAFTcheckpoint=ours2024.03 | 7.51 | — | — | — | — | — | — | |
| DKT-IGEVcheckpoint=ours2024.03 | 7.53 | — | — | — | — | — | — | |
| IGEV++zero-shot=true2026.05 | 7.8 | — | — | — | — | — | — | |
| IGEV++2026.06 | 7.8 | — | 5.72 | — | — | — | — | |
| LoS2025.01 | 8.03 | — | 1.75 | — | — | 8.78 | — | |
| DSMNetZero-shot generalization=true, Trained on=SceneFlow2024.03 | 8.1 | — | — | — | — | — | — | |
| SMFormer2026.04 | 8.1 | — | — | — | — | — | — | |
| Former-RAFT-DAMzero-shot=true2026.05 | 8.1 | — | — | — | — | — | — | |
| Former-RAFT-DAM2026.06 | 8.1 | — | 5.1 | — | — | — | — | |
| CREStereo2025.01 | 8.13 | — | 2.1 | — | — | 10.5 | — | |
| IGEV-Stereo2025.01 | 8.16 | — | 3.64 | — | — | 15.1 | — | |
| DLNRcheckpoint=provided by authors2024.03 | 8.21 | — | — | — | — | — | — | |
| IGEV-StereoZero-shot generalization=true, Trained on=SceneFlow2024.03 | 8.8 | — | — | — | — | — | — | |
| IGEVzero-shot=true2026.05 | 8.8 | — | — | — | — | — | — | |
| IGEV2026.06 | 8.8 | — | 5.92 | — | — | — | — | |
| Selective-IGEVzero-shot=true2026.05 | 9.2 | — | — | — | — | — | — | |
| Selective-IGEV2026.06 | 9.2 | — | 6.25 | — | — | — | — | |
| RAFT-Stereo2025.01 | 9.37 | — | 2.71 | — | — | 12.6 | — | |
| RAFT-StereoZero-shot generalization=true, Trained on=SceneFlow2024.03 | 9.4 | — | — | — | — | — | — | |
| Former-PSMNet (SAM)Pre-trained VFM=true2026.04 | 9.4 | — | — | — | — | — | — | |
| HVT-RAFTzero-shot=true2026.05 | 10.4 | — | — | — | — | — | — | |
| HVT-RAFT2026.06 | 10.4 | — | 5.58 | — | — | — | — | |
| CroCo-Stereo2025.01 | 11.1 | — | 2.36 | — | — | 10.6 | — | |
| GMStereo2025.01 | 11.7 | — | 1.89 | — | — | 8.03 | — | |
| RAFT-Stereocheckpoint=provided by authors2024.03 | 11.78 | — | — | — | — | — | — | |
| IGEV-Stereocheckpoint=provided by authors2024.03 | 11.87 | — | — | — | — | — | — | |
| RAFT-Stereo2026.04 | 12.6 | — | — | — | — | — | — | |
| RAFT-Stereozero-shot=true2026.05 | 12.6 | — | — | — | — | — | — | |
| RAFT-Stereo2026.06 | 12.6 | — | 6.52 | — | — | — | — | |
| HITNet2025.01 | 12.8 | — | 3.29 | — | — | 14.5 | — | |
| URS-StereoZero-shot=true, Training Dataset=SceneFlow2026.07 | 13.03 | — | 1.71 | — | — | — | — | |
| Lite Any StereoZero-shot=true, Training Dataset=SceneFlow2026.07 | 13.13 | — | 1.6 | — | — | — | — | |
| Mask-CFNetzero-shot=true2026.05 | 13.7 | — | — | — | — | — | — | |
| Mask-CFNet2026.06 | 13.7 | — | 7.5 | — | — | — | — | |
| DSMNetzero-shot=true2026.05 | 13.8 | — | — | — | — | — | — | |
| DSMNet2026.06 | 13.8 | — | 8.18 | — | — | — | — | |
| CREStereo++zero-shot=true2026.05 | 14.8 | — | — | — | — | — | — | |
| CREStereo++2026.06 | 14.8 | — | 7.28 | — | — | — | — | |
| Lite-CREStereo++Zero-shot=true, Training Dataset=SceneFlow2026.07 | 14.91 | — | 3.32 | — | — | — | — | |
| CFNet2026.04 | 15.4 | — | — | — | — | — | — | |
| PCWNetcheckpoint=provided by authors2024.03 | 15.55 | — | — | — | — | — | — | |
| PCWNet2026.04 | 15.8 | — | — | — | — | — | — | |
| CFNetadaptation=none, Robust Vision Challenge 2020 context=true2021.04 | 16.1 | 26.2 | 5.07 | 2 | — | — | — | |
| CFNetadaptation=none, top methods in the past two years context=true2021.04 | 16.1 | 26.2 | 5.07 | 1 | — | — | — | |
| NMRF-SwinT2026.06 | 16.36 | — | 13.99 | — | — | — | — | |
| NLCANet_V2adaptation=none, Robust Vision Challenge 2020 context=true2021.04 | 16.4 | 29.4 | 5.6 | 3 | — | — | — | |
| HSMNetadaptation=none, Robust Vision Challenge 2020 context=true2021.04 | 16.5 | 31.2 | 3.44 | 1 | — | — | — | |
| LightStereo-MZero-shot=true, Training Dataset=SceneFlow2026.07 | 16.99 | — | 2.06 | — | — | — | — | |
| LightStereo-LZero-shot=true, Training Dataset=SceneFlow2026.07 | 17.23 | — | 2.88 | — | — | — | — | |
| GWCNetcheckpoint=provided by authors2024.03 | 17.83 | — | — | — | — | — | — | |
| GANetcheckpoint=provided by authors2024.03 | 18.79 | — | — | — | — | — | — | |
| LoS2026.04 | 19.6 | — | — | — | — | — | — | |
| FastACVZero-shot=true, Training Dataset=SceneFlow2026.07 | 19.61 | — | 4.66 | — | — | — | — | |
| PSMNetcheckpoint=provided by authors2024.03 | 21.85 | — | — | — | — | — | — | |
| CGF-ACVcheckpoint=provided by authors2024.03 | 23.76 | — | — | — | — | — | — | |
| GANetadaptation=none, Robust Vision Challenge 2020 context=true2021.04 | 24.9 | 43.1 | 15.8 | 7 | — | — | — | |
| MobileStereoNet-3DZero-shot=true, Training Dataset=SceneFlow2026.07 | 25.26 | — | 4.61 | — | — | — | — | |
| UCFNet_pretrain2026.04 | 26 | — | — | — | — | — | — | |
| CoEXZero-shot=true, Training Dataset=SceneFlow2026.07 | 26.42 | — | 4.9 | — | — | — | — | |
| BANet-2DZero-shot=true, Training Dataset=SceneFlow2026.07 | 26.79 | — | 6.96 | — | — | — | — | |
| FastACV+Zero-shot=true, Training Dataset=SceneFlow2026.07 | 27.34 | — | 7.16 | — | — | — | — | |
| BANet-3DZero-shot=true, Training Dataset=SceneFlow2026.07 | 28.79 | — | 8.05 | — | — | — | — | |
| iResNetadaptation=none, top methods in the past two years context=true2021.04 | 31.7 | 45.9 | 6.56 | 2 | — | — | — | |
| AANetadaptation=none, Robust Vision Challenge 2020 context=true2021.04 | 31.8 | 42.9 | 12.8 | 6 | — | — | — | |
| Deeppruneradaptation=none, top methods in the past two years context=true2021.04 | 36.4 | 57.1 | 6.56 | 3 | — | — | — | |
| MobileStereoNet-2DZero-shot=true, Training Dataset=SceneFlow2026.07 | 37.98 | — | 7.54 | — | — | — | — | |
| CVANetadaptation=none, Robust Vision Challenge 2020 context=true2021.04 | 38.5 | 58.5 | 8.64 | 5 | — | — | — | |
| AIO-Stereo2025.10 | — | 6.08 | 0.85 | — | — | 6.78 | 24 | |
| BaseNetasymmetric factor=8, degradation=BIC2022.04 | — | — | 2.049 | — | 15.33 | — | — | |
| BaseNet+AEasymmetric factor=8, degradation=BIC2022.04 | — | — | 2.02 | — | 14.3 | — | — | |
| BaseNet+CLasymmetric factor=8, degradation=BIC2022.04 | — | — | 2.129 | — | 16.51 | — | — | |
| CFNetEvaluation Protocol=Zero-shot, Weights Source=Authors' weights2024.03 | — | — | 15.77 | — | — | — | — | |
| CFNetEvaluation Protocol=Fine-tuned, Weights Source=Reproduced from online submission setting2024.03 | — | — | 18.42 | — | — | — | — | |
| DEFOM-Stereo2025.10 | — | 5.81 | 0.79 | — | — | 5.81 | 25.2 | |
| DKT-IGEVEvaluation Protocol=Fine-tuned, Framework=DKT2024.03 | — | — | 6.94 | — | — | — | — | |
| DKT-RAFTEvaluation Protocol=Fine-tuned, Framework=DKT2024.03 | — | — | 5.94 | — | — | — | — | |
| feature-metric consistency (Ours)asymmetric factor=8, degradation=BIC2022.04 | — | — | 1.584 | — | 9.9 | — | — | |
| FoundationStereo2025.10 | — | 4.39 | 0.78 | — | — | 6.48 | 22.5 | |
| MatchAttentionXL2025.10 | — | 3.49 | 0.68 | — | — | 5.66 | 21.3 | |
| MGS-Stereo2025.10 | — | 5.69 | 0.74 | — | — | 5.91 | 23.4 | |
| PCVNetEvaluation Protocol=Fine-tuned, Training Data=Retrained with only Booster training set2024.03 | — | — | 22.88 | — | — | — | — | |
| RAFT-StereoEvaluation Protocol=Zero-shot, Weights Source=Authors' weights2024.03 | — | — | 8.41 | — | — | — | — | |
| RAFT-StereoEvaluation Protocol=Fine-tuned, Weights Source=Reproduced from online submission setting2024.03 | — | — | 15.98 | — | — | — | — |