Monocular Metric Depth Estimation on ScanNet++
97.6δ1VLM3-4b
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| VLM3-4bModel Type=VLM Trained on Metric Depth Estimation2026.05 | 97.6 | — | — | |
| UniDACBackbone=ViT-L, Zero-shot evaluation=true, Training domain=Universal2026.03 | 91.8 | 0.097 | 0.277 | |
| DACIBackbone=Swin-L, Zero-shot evaluation=true, Training domain=Indoor2026.03 | 85.4 | 0.128 | 0.287 | |
| DACIBackbone=ResNet101, Zero-shot evaluation=true, Training domain=Indoor2026.03 | 85.2 | 0.132 | 0.309 | |
| DepthLM-7bModel Type=VLM Trained on Metric Depth Estimation2026.05 | 85 | — | — | |
| DepthLM-3bModel Type=VLM Trained on Metric Depth Estimation2026.05 | 83.8 | — | — | |
| DACUBackbone=Swin-L, Zero-shot evaluation=true, Training domain=Universal2026.03 | 65.8 | 0.233 | 0.464 | |
| Seed1.5-VLModel Type=VLM Trained on Metric Depth Estimation2026.05 | 59.3 | — | — | |
| GPT-5Model Type=VLM2026.05 | 42.8 | — | — | |
| Gemini-2.5-ProModel Type=VLM2026.05 | 38 | — | — | |
| Qwen3-VL-32bModel Type=VLM2026.05 | 37.3 | — | — | |
| SpatialRGPT-8BModel Type=Spatial VLM2026.05 | 34.6 | — | — | |
| Qwen2.5-VL-72bModel Type=VLM2026.05 | 27.2 | — | — | |
| SpaceLLaVA-13BModel Type=Spatial VLM2026.05 | 26.9 | — | — | |
| DACOBackbone=ResNet101, Zero-shot evaluation=true, Training domain=Outdoor2026.03 | 25.6 | 0.901 | 2.312 | |
| Qwen3-VL-4bModel Type=VLM2026.05 | 14.7 | — | — | |
| DACOBackbone=Swin-L, Zero-shot evaluation=true, Training domain=Outdoor2026.03 | 10.9 | 1.412 | 3.539 |