Spatial Reasoning on ViewBench-Text (5–15% overlap)
1AccuracyTarget View (Oracle)
Evaluation Results
| Method | Links | |
|---|---|---|
| Target View (Oracle)Depth=GT2026.04 | 1 | |
| Token Warping (Backward-Adaptive)Depth=GT2026.04 | 0.7789 | |
| Token Warping (Backward-Nearest)Depth=Pred.2026.04 | 0.7538 | |
| Token Warping (Backward-Nearest)Depth=GT2026.04 | 0.7487 | |
| Token Warping (Backward-Adaptive)Depth=Pred.2026.04 | 0.7337 | |
| Pixel-Wise Warping (Backward)Depth=GT2026.04 | 0.7186 | |
| Pixel-Wise Warping (Forward)Depth=GT2026.04 | 0.7085 | |
| SpatialLadderDepth=GT2026.04 | 0.7035 | |
| GenWarpDepth=GT2026.04 | 0.6935 | |
| Pixel-Wise Warping (Forward)Depth=Pred.2026.04 | 0.6935 | |
| SpaceQwenDepth=GT2026.04 | 0.6834 | |
| SpaceR-SFT-7BDepth=GT2026.04 | 0.6784 | |
| Pixel-Wise Warping (Backward)Depth=Pred.2026.04 | 0.6784 | |
| Token Warping (Forward)Depth=Pred.2026.04 | 0.6683 | |
| VLM-3RDepth=GT2026.04 | 0.6382 | |
| Token Warping (Forward)Depth=GT2026.04 | 0.603 | |
| MindCubeDepth=GT2026.04 | 0.593 | |
| InternVL3Depth=GT2026.04 | 0.5628 | |
| Kimi-VL-ThinkingDepth=GT2026.04 | 0.4925 | |
| VeBrainDepth=GT2026.04 | 0.4925 | |
| SpaceThinkerDepth=GT2026.04 | 0.4874 | |
| LLaVA-OneVision-1.5Depth=GT2026.04 | 0.4824 | |
| SpatialReasonerDepth=GT2026.04 | 0.4673 | |
| Qwen2.5-VLDepth=GT2026.04 | 0.4623 | |
| ViLaSRDepth=GT2026.04 | 0.4422 | |
| VST-SFTDepth=GT2026.04 | 0.4271 | |
| Qwen3-VLDepth=GT2026.04 | 0.4171 | |
| RoboBrain-2.0Depth=GT2026.04 | 0.3769 | |
| VST-RLDepth=GT2026.04 | 0.2814 | |
| Cambrian-1Depth=GT2026.04 | 0.0905 | |
| VG-LLMDepth=GT2026.04 | 0.0593 |