Monocular Depth Estimation on nuScenes (val)
0.08Abs RelDriveTok
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| DriveTokInput=6 × 256 × 704, latency=383.53 ms, Peak Memory=8562.64 MB2026.03 | 0.08 | — | — | — | 93 | — | — | — | |
| DepthAnythingEvaluation protocol=finetuned, Params.=335.79 M, Latency=130.41 ms2025.12 | 0.107 | 0.831 | 3.866 | 0.185 | 89 | 95 | 97.4 | — | |
| MonoViT + MB + OursLearning paradigm=Supervised, Strategy=Matrix Binning + CKD/VRKD, Params.=28.31 M, Latency=25.40 ms2025.12 | 0.134 | 1.072 | 4.096 | 0.214 | 83 | 91.4 | 95.4 | 15 | |
| HRDepth + MB + OursLearning paradigm=Supervised, Strategy=Matrix Binning + CKD/VRKD, Params.=14.80 M, Latency=13.92 ms2025.12 | 0.156 | 1.097 | 5.02 | 0.244 | 79.2 | 91.1 | 95.8 | 7.18 | |
| Monodepth2 + MB + OursLearning paradigm=Supervised, Strategy=Matrix Binning + CKD/VRKD, Params.=15.03 M, Latency=11.71 ms2025.12 | 0.158 | 1.135 | 4.915 | 0.243 | 80.5 | 91.4 | 95.7 | 13.42 | |
| MonoViT + MBLearning paradigm=Supervised, Strategy=Matrix Binning, Params.=28.31 M, Latency=25.40 ms2025.12 | 0.169 | 1.616 | 5.121 | 0.252 | 77 | 88.4 | 94 | 0.54 | |
| MonoViTLearning paradigm=Supervised, Params.=27.87 M, Latency=17.26 ms2025.12 | 0.17 | 1.643 | 5.174 | 0.253 | 76.9 | 88.4 | 94 | 0 | |
| HRDepthLearning paradigm=Supervised, Params.=14.10 M, Latency=5.67 ms2025.12 | 0.172 | 1.313 | 5.523 | 0.267 | 76.5 | 89.3 | 94.6 | 0 | |
| HRDepth + MBLearning paradigm=Supervised, Strategy=Matrix Binning, Params.=14.80 M, Latency=13.92 ms2025.12 | 0.174 | 1.36 | 5.439 | 0.278 | 76.2 | 88.1 | 93.4 | -1.48 | |
| Monodepth2Learning paradigm=Supervised, Params.=14.33 M, Latency=3.44 ms2025.12 | 0.182 | 1.593 | 5.961 | 0.295 | 74.4 | 86.8 | 92.6 | 0 | |
| Monodepth2 + MBLearning paradigm=Supervised, Strategy=Matrix Binning, Params.=15.03 M, Latency=11.71 ms2025.12 | 0.188 | 1.719 | 6.145 | 0.311 | 73.3 | 85.7 | 91.7 | -3.35 | |
| DepthAnythingEvaluation protocol=zero-shot, Params.=335.79 M, Latency=130.41 ms2025.12 | 0.208 | 1.324 | 5.108 | 0.28 | 73.5 | 90.1 | 95.1 | — | |
| EGA-Depth-MRLearning paradigm=Self-supervised, Params.=not public, Latency=not public2025.12 | 0.228 | 2.113 | 6.738 | 0.311 | 72.8 | 88.5 | 94 | — | |
| OmniNWMCategory=Generative2025.10 | 0.23 | — | — | — | 81 | 93 | — | — | |
| OmniNWMCategory=Generative2025.10 | 0.23 | — | — | — | 81 | 93 | — | — | |
| EGA-Depth-LRLearning paradigm=Self-supervised, Params.=not public, Latency=not public2025.12 | 0.239 | 2.357 | 6.801 | 0.317 | 72.3 | 88 | 93.6 | — | |
| SurroundDepthLearning paradigm=Self-supervised, Params.=59.56 M, Latency=10.82 ms2025.12 | 0.245 | 3.067 | 6.835 | 0.321 | 71.9 | 87.8 | 93.5 | — | |
| DepthProInput=256 × 704, latency=5179.93 ms, Peak Memory=30559.55 MB2026.03 | 0.26 | — | — | — | 47 | — | — | — | |
| M2DepthCategory=Discriminative2025.10 | 0.26 | — | — | — | 73 | 87 | — | — | |
| M²DepthCategory=Discriminative2025.10 | 0.26 | — | — | — | 73 | 87 | — | — | |
| Metric3D-V2Input=256 × 704, latency=403.49 ms, Peak Memory=2982.53 MB2026.03 | 0.28 | — | — | — | 69 | — | — | — | |
| SurroundDepthCategory=Discriminative2025.10 | 0.28 | — | — | — | 66 | 84 | — | — | |
| SurroundDepthCategory=Discriminative2025.10 | 0.28 | — | — | — | 66 | 84 | — | — | |
| Monodepth2Learning paradigm=Self-supervised, Params.=14.33 M, Latency=3.44 ms2025.12 | 0.287 | 3.349 | 7.184 | 0.345 | 64.1 | 84.5 | 92.5 | — | |
| FSMLearning paradigm=Self-supervised, Params.=14.33 M, Latency=3.44 ms2025.12 | 0.299 | — | — | — | — | — | — | — | |
| PackNet-SfMLearning paradigm=Self-supervised, Params.=128.29 M, Latency=57.86 ms2025.12 | 0.309 | 2.891 | 7.994 | 0.39 | 54.7 | 79.6 | 89.9 | — | |
| Dist-4DCategory=Generative2025.10 | 0.39 | — | — | — | 58 | 81 | — | — | |
| Dist-4DCategory=Generative2025.10 | 0.39 | — | — | — | 58 | 81 | — | — | |
| UniDepthV2-LInput=256 × 704, latency=100.45 ms, Peak Memory=3902.32 MB2026.03 | 0.4 | — | — | — | 31 | — | — | — | |
| UniDepthV2-BInput=256 × 704, latency=62.45 ms, Peak Memory=2194.65 MB2026.03 | 0.52 | — | — | — | 30 | — | — | — |