# Stereo Matching on Middlebury v3 Source: [Stereo Matching on Middlebury v3]() Collection index: [llms.txt]() Data updated: 2026-06-30T09:25:58.788Z Comprehensive evaluation of Stereo Matching performance on the Middlebury v3 dataset, reporting timing metrics and detailed error statistics (Bad Pixel Rates, Average Error, RMS Error) for both Non-Occluded and All pixels. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Stereo Matching | | Dataset (including any stated split) | Middlebury v3 | | Metric | Bad Pixel Rate (Thresh 2.0) | | Unit | PERCENT | | Direction | Lower is better | | Results loaded | 48 of 48 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Bad Pixel Rate (Thresh 2.0)](): Lower is better; unit: PERCENT. - [Time (s)](): Lower is better; unit: TIME. - [Time/MP (s)](): Lower is better; unit: TIME. - [Time/GD (s)](): Lower is better; unit: TIME. - [NonOcc Bad Pixel Rate (Thresh 2.0)](): Lower is better; unit: PERCENT. - [NonOcc Bad Pixel Rate (Thresh 4.0)](): Lower is better; unit: PERCENT. - [NonOcc Average Error](): Lower is better; unit: SCALAR. - [NonOcc RMS Error](): Lower is better; unit: SCALAR. - [NonOcc Error Percentile (90th)](): Lower is better; unit: SCALAR. - [NonOcc Error Percentile (99th)](): Lower is better; unit: SCALAR. - [All Bad Pixel Rate (Thresh 2.0)](): Lower is better; unit: PERCENT. - [All Bad Pixel Rate (Thresh 4.0)](): Lower is better; unit: PERCENT. - [All Average Error](): Lower is better; unit: SCALAR. - [All RMS Error](): Lower is better; unit: SCALAR. - [All Error Percentile (90th)](): Lower is better; unit: SCALAR. - [All Error Percentile (99th)](): Lower is better; unit: SCALAR. - [RMS Error](): Lower is better; unit: SCALAR. - [Average Error](): Lower is better; unit: SCALAR. - [Bad Pixel Rate (Thresh 0.5)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate (Thresh 1.0)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate (Thresh 4.0)](): Lower is better; unit: PERCENT. - [Success Rate (0.5)](): Higher is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Avg)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Aust)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (AustP)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Bcyc)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Clas)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (ClaE)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Comp)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Crus)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (CruP)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Djmb)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (DjmL)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Hoop)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Lvgr)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Nkub)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Plnt)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 0.5% (Stair)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Avg)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Aust)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (AustP)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Bcyc)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Clas)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (ClaE)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Comp)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Crus)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (CruP)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Djmb)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (DjmL)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Hoop)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Lvgr)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Nkub)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Plnt)](): Lower is better; unit: PERCENT. - [Bad Pixel Rate 1.0% (Stair)](): Lower is better; unit: PERCENT. ## Recorded results | Method | Bad Pixel Rate (Thresh 2.0) | Result date | Report paper | | --- | --- | --- | --- | | MatchAttentionXL | 2.08 | Not recorded | 68fa7c351e209131cea8d08e | | MatchAttentionRT | 4.38 | Not recorded | 68fa7c351e209131cea8d08e | | NOSS-ROB | 5.01 | Not recorded | 68fa6e661e209131cea3fb71 | | LocalExp | 5.43 | Not recorded | 68fa6e661e209131cea3fb71 | | CRLE | 5.75 | Not recorded | 68fa6e661e209131cea3fb71 | | HITNet | 6.46 | Not recorded | 68fa6e661e209131cea3fb71 | | DEFOM-Stereo\_RVC | 6.9 | Not recorded | 68fa783c1e209131cea77578 | | LEAStereo | 7.15 | Not recorded | 68fa6e661e209131cea3fb71 | | Fast-FSD-416 | 7.2 | Not recorded | 68fa7c351e209131cea8d08e | | Monster++-416 | 7.36 | Not recorded | 68fa7c351e209131cea8d08e | | MC-CNN | 8.08 | Not recorded | 68fa6e661e209131cea3fb71 | | Zero-RAFT-Stereo | 8.43 | Not recorded | 68fa7c351e209131cea8d08e | | LoS\_RVC | 9.3 | Not recorded | 68fa783c1e209131cea77578 | | CREStereo++\_RVC | 9.46 | Not recorded | 68fa783c1e209131cea77578 | | HSM | 10.2 | Not recorded | 68fa6e661e209131cea3fb71 | | MatchAttentionRT | 11.29 | Not recorded | 68fa7c351e209131cea8d08e | | iRaftStereo\_RVC | 13.3 | Not recorded | 68fa783c1e209131cea77578 | | RT-Monster++-416 | 13.61 | Not recorded | 68fa7c351e209131cea8d08e | | CFNet\_RVC | 16.1 | Not recorded | 68fa783c1e209131cea77578 | | NLCA\_NET\_v2\_RVC | 16.4 | Not recorded | 68fa783c1e209131cea77578 | | UCFNet\_RVC | 16.7 | Not recorded | 68fa783c1e209131cea77578 | | RT-IGEV-416 | 18.27 | Not recorded | 68fa7c351e209131cea8d08e | | EdgeStereo | 18.7 | Not recorded | 68fa6e661e209131cea3fb71 | | Fast-FSD-192 | 24.61 | Not recorded | 68fa7c351e209131cea8d08e | | RT-Monster++-192 | 26.03 | Not recorded | 68fa7c351e209131cea8d08e | | DN-CSS\_ROB | 28.3 | Not recorded | 68fa783c1e209131cea77578 | | LiteAnyStereo | 30.77 | Not recorded | 68fa7c351e209131cea8d08e | | iResNet\_ROB | 31.7 | Not recorded | 68fa783c1e209131cea77578 | | RT-IGEV-192 | 32.37 | Not recorded | 68fa7c351e209131cea8d08e | | LiteAnyStereo | 35.77 | Not recorded | 68fa7c351e209131cea8d08e | | IINet | 36.38 | Not recorded | 68fa7c351e209131cea8d08e | | BANet-3D-416 | 48.96 | Not recorded | 68fa7c351e209131cea8d08e | | LightStereo-L | 51.49 | Not recorded | 68fa7c351e209131cea8d08e | | BANet-3D-192 | 58.34 | Not recorded | 68fa7c351e209131cea8d08e | | BANet-2D | 58.54 | Not recorded | 68fa7c351e209131cea8d08e | ### MatchAttentionXL Bad Pixel Rate (Thresh 2.0): 2.08. Result ID: 6a438b8cfbc373592c958dce. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Million-scale","FLOPs (G)":"5240.2","Latency (NX-16)":"1287.8","Latency (AGX-64)":"493.1","Latency (4060Ti)":"248.8","Params (M)":"507.1"}. This record does not assert independent reproduction. ### MatchAttentionRT Bad Pixel Rate (Thresh 2.0): 4.38. Result ID: 6a438b8cfbc373592c958dcb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Million-scale","FLOPs (G)":"62.5","Latency (NX-16)":"79.1","Latency (AGX-64)":"36.7","Latency (4060Ti)":"9.3","Params (M)":"10.92"}. This record does not assert independent reproduction. ### NOSS-ROB Bad Pixel Rate (Thresh 2.0): 5.01. Result ID: 699030343a6c314849fec3c9. [Reporting paper]() Recorded evaluation configuration: {"Run-time":"662s (CPU)"}. This record does not assert independent reproduction. ### LocalExp Bad Pixel Rate (Thresh 2.0): 5.43. Result ID: 699030343a6c314849fec3ca. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Run-time":"881s (CPU)"}. This record does not assert independent reproduction. ### CRLE Bad Pixel Rate (Thresh 2.0): 5.75. Result ID: 699030343a6c314849fec3cb. [Reporting paper]() Recorded evaluation configuration: {"Run-time":"1589s (CPU)"}. This record does not assert independent reproduction. ### HITNet Bad Pixel Rate (Thresh 2.0): 6.46. Result ID: 699030343a6c314849fec3c7. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Run-time":"0.14 s"}. This record does not assert independent reproduction. ### DEFOM-Stereo\_RVC Bad Pixel Rate (Thresh 2.0): 6.9. Result ID: 699581363badfaffe6794d4f. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Evaluation protocol":"Recent RVC Model"}. This record does not assert independent reproduction. ### LEAStereo Bad Pixel Rate (Thresh 2.0): 7.15. Result ID: 699030343a6c314849fec3c8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Run-time":"2.9 s"}. This record does not assert independent reproduction. ### Fast-FSD-416 Bad Pixel Rate (Thresh 2.0): 7.2. Result ID: 6a438b8cfbc373592c958dc7. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Training Data":"Million-scale","Max Disparity":"416","FLOPs (G)":"1811.6","Latency (NX-16)":"1315.0","Latency (AGX-64)":"497.7","Latency (4060Ti)":"421.0","Params (M)":"17.65"}. This record does not assert independent reproduction. ### Monster++-416 Bad Pixel Rate (Thresh 2.0): 7.36. Result ID: 6a438b8cfbc373592c958dcd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Million-scale","Max Disparity":"416","FLOPs (G)":"10843.0","Latency (NX-16)":"3581.7","Latency (AGX-64)":"1493.6","Latency (4060Ti)":"567.1","Params (M)":"388.7"}. This record does not assert independent reproduction. ### MC-CNN Bad Pixel Rate (Thresh 2.0): 8.08. Result ID: 699030343a6c314849fec3cd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Run-time":"150 s"}. This record does not assert independent reproduction. ### Zero-RAFT-Stereo Bad Pixel Rate (Thresh 2.0): 8.43. Result ID: 6a438b8cfbc373592c958dcc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Million-scale","FLOPs (G)":"6384.5","Latency (NX-16)":"1007.6","Latency (AGX-64)":"389.2","Latency (4060Ti)":"216.1","Params (M)":"11.12"}. This record does not assert independent reproduction. ### LoS\_RVC Bad Pixel Rate (Thresh 2.0): 9.3. Result ID: 699581363badfaffe6794d4e. [Reporting paper]() Recorded evaluation configuration: {"Evaluation protocol":"Recent RVC Model"}. This record does not assert independent reproduction. ### CREStereo++\_RVC Bad Pixel Rate (Thresh 2.0): 9.46. Result ID: 699581363badfaffe6794d4c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Evaluation protocol":"Robust Vision Challenge 2022"}. This record does not assert independent reproduction. ### HSM Bad Pixel Rate (Thresh 2.0): 10.2. Result ID: 699030343a6c314849fec3cc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Run-time":"0.51 s"}. This record does not assert independent reproduction. ### MatchAttentionRT Bad Pixel Rate (Thresh 2.0): 11.29. Result ID: 6a438b8cfbc373592c958dd7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Scene Flow","FLOPs (G)":"62.5","Latency (NX-16)":"79.1","Latency (AGX-64)":"36.7","Latency (4060Ti)":"9.3","Params (M)":"10.92"}. This record does not assert independent reproduction. ### iRaftStereo\_RVC Bad Pixel Rate (Thresh 2.0): 13.3. Result ID: 699581363badfaffe6794d4b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Evaluation protocol":"Robust Vision Challenge 2022"}. This record does not assert independent reproduction. ### RT-Monster++-416 Bad Pixel Rate (Thresh 2.0): 13.61. Result ID: 6a438b8cfbc373592c958dc9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Million-scale","Max Disparity":"416","FLOPs (G)":"579.1","Latency (NX-16)":"295.0","Latency (AGX-64)":"121.2","Latency (4060Ti)":"44.8","Params (M)":"42.82"}. This record does not assert independent reproduction. ### CFNet\_RVC Bad Pixel Rate (Thresh 2.0): 16.1. Result ID: 699581363badfaffe6794d4a. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Evaluation protocol":"Robust Vision Challenge 2020"}. This record does not assert independent reproduction. ### NLCA\_NET\_v2\_RVC Bad Pixel Rate (Thresh 2.0): 16.4. Result ID: 699581363badfaffe6794d49. [Reporting paper]() Recorded evaluation configuration: {"Evaluation protocol":"Robust Vision Challenge 2020"}. This record does not assert independent reproduction. ### UCFNet\_RVC Bad Pixel Rate (Thresh 2.0): 16.7. Result ID: 699581363badfaffe6794d4d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Evaluation protocol":"Recent RVC Model"}. This record does not assert independent reproduction. ### RT-IGEV-416 Bad Pixel Rate (Thresh 2.0): 18.27. Result ID: 6a438b8cfbc373592c958dd2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Scene Flow","Max Disparity":"416","FLOPs (G)":"573.9","Latency (NX-16)":"927.1","Latency (AGX-64)":"335.4","Latency (4060Ti)":"117.1","Params (M)":"4.17"}. This record does not assert independent reproduction. ### EdgeStereo Bad Pixel Rate (Thresh 2.0): 18.7. Result ID: 699030343a6c314849fec3ce. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Run-time":"0.35 s"}. This record does not assert independent reproduction. ### Fast-FSD-192 Bad Pixel Rate (Thresh 2.0): 24.61. Result ID: 6a438b8cfbc373592c958dc6. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Training Data":"Million-scale","Max Disparity":"192","FLOPs (G)":"1145.0","Latency (NX-16)":"777.5","Latency (AGX-64)":"299.1","Latency (4060Ti)":"222.2","Params (M)":"17.65"}. This record does not assert independent reproduction. ### RT-Monster++-192 Bad Pixel Rate (Thresh 2.0): 26.03. Result ID: 6a438b8cfbc373592c958dc8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Million-scale","Max Disparity":"192","FLOPs (G)":"578.9","Latency (NX-16)":"294.8","Latency (AGX-64)":"118.2","Latency (4060Ti)":"44.8","Params (M)":"42.82"}. This record does not assert independent reproduction. ### DN-CSS\_ROB Bad Pixel Rate (Thresh 2.0): 28.3. Result ID: 699581363badfaffe6794d47. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Evaluation protocol":"Robust Vision Challenge 2018"}. This record does not assert independent reproduction. ### LiteAnyStereo Bad Pixel Rate (Thresh 2.0): 30.77. Result ID: 6a438b8cfbc373592c958dca. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Million-scale","FLOPs (G)":"61.1","Latency (NX-16)":"133.7","Latency (AGX-64)":"60.9","Latency (4060Ti)":"16.6","Params (M)":"7.60"}. This record does not assert independent reproduction. ### iResNet\_ROB Bad Pixel Rate (Thresh 2.0): 31.7. Result ID: 699581363badfaffe6794d48. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Evaluation protocol":"Robust Vision Challenge 2018"}. This record does not assert independent reproduction. ### RT-IGEV-192 Bad Pixel Rate (Thresh 2.0): 32.37. Result ID: 6a438b8cfbc373592c958dd1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Scene Flow","Max Disparity":"192","FLOPs (G)":"549.0","Latency (NX-16)":"346.7","Latency (AGX-64)":"143.3","Latency (4060Ti)":"50.4","Params (M)":"4.17"}. This record does not assert independent reproduction. ### LiteAnyStereo Bad Pixel Rate (Thresh 2.0): 35.77. Result ID: 6a438b8cfbc373592c958dd6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Scene Flow","FLOPs (G)":"61.1","Latency (NX-16)":"133.7","Latency (AGX-64)":"60.9","Latency (4060Ti)":"16.6","Params (M)":"7.60"}. This record does not assert independent reproduction. ### IINet Bad Pixel Rate (Thresh 2.0): 36.38. Result ID: 6a438b8cfbc373592c958dcf. [Reporting paper]() Recorded evaluation configuration: {"Training Data":"Scene Flow","FLOPs (G)":"185.4","Latency (NX-16)":"101.5","Latency (AGX-64)":"44.5","Latency (4060Ti)":"13.7","Params (M)":"19.56"}. This record does not assert independent reproduction. ### BANet-3D-416 Bad Pixel Rate (Thresh 2.0): 48.96. Result ID: 6a438b8cfbc373592c958dd5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Scene Flow","Max Disparity":"416","FLOPs (G)":"171.4","Latency (NX-16)":"248.7","Latency (AGX-64)":"108.5","Latency (4060Ti)":"71.5","Params (M)":"3.63"}. This record does not assert independent reproduction. ### LightStereo-L Bad Pixel Rate (Thresh 2.0): 51.49. Result ID: 6a438b8cfbc373592c958dd0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Scene Flow","FLOPs (G)":"168.4","Latency (NX-16)":"154.3","Latency (AGX-64)":"71.1","Latency (4060Ti)":"20.4","Params (M)":"24.29"}. This record does not assert independent reproduction. ### BANet-3D-192 Bad Pixel Rate (Thresh 2.0): 58.34. Result ID: 6a438b8cfbc373592c958dd4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Scene Flow","Max Disparity":"192","FLOPs (G)":"116.0","Latency (NX-16)":"150.1","Latency (AGX-64)":"66.8","Latency (4060Ti)":"35.4","Params (M)":"3.63"}. This record does not assert independent reproduction. ### BANet-2D Bad Pixel Rate (Thresh 2.0): 58.54. Result ID: 6a438b8cfbc373592c958dd3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Data":"Scene Flow","FLOPs (G)":"65.1","Latency (NX-16)":"76.6","Latency (AGX-64)":"47.8","Latency (4060Ti)":"8.0","Params (M)":"5.46"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.