Research map / 3D vision / Depth and stereo
Last-Layer-Centric Feature Recombination: Unleashing 3D Geometric Knowledge in DINOv3 for Monocular Depth Estimation
Gongshu Wang, Zhirui Wang, Kan Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add LFR module | DINOv3-B · NYU Depth v2 | AbsRel: 0.072 -> 0.070 | Improved |
| add LFR module | DINOv3-L · NYU Depth v2 | AbsRel: 0.060 -> 0.057 | Improved |
| add LFR module | DINOv3-B · KITTI (Eigen split) | AbsRel: 0.057 -> 0.049 | Improved |
| add LFR module | DINOv3-L · KITTI (Eigen split) | AbsRel: 0.046 -> 0.043 | Improved |
| add LFR module | DINOv3-L · SUN RGB-D (zero-shot from NYU) | AbsRel: 0.101 -> 0.099 | Improved |
| add LFR module | DINOv3-L · Argoverse (zero-shot from KITTI) | AbsRel: 0.219 -> 0.204 | Improved |
| use last layer only instead of uniformly sampled layers (ablation) | DINOv3-L + DPT · KITTI (Eigen split, ablation) | AbsRel: 0.046 -> 0.045 | Improved |
| add recombination of selected intermediate layers (ablation) | DINOv3-L + DPT · KITTI (Eigen split, ablation) | AbsRel: 0.045 -> 0.043 | Improved |
| add recombination of selected intermediate layers (ablation) | DINOv3-L + DPT · KITTI (Eigen split, ablation) | RMSE: 1.743 -> 1.721 | Improved |
| add hierarchical normalization loss (ablation) | DINOv3-L + LFR · KITTI (Eigen split, ablation) | SIlog: 0.060 -> 0.059 | Improved |
| add multi-level weighted prediction (ablation) | DINOv3-L + LFR · KITTI (Eigen split, ablation) | RMSE: 1.719 -> 1.711 | Improved |
| select auxiliary layers by minimal similarity instead of predicted scores (ablation) | LFR · KITTI (Eigen split, ablation) | AbsRel: 0.044 -> 0.043 | Improved |
Closest papers by meaning
- Geometric Distillation from Rectified Stereo: Leveraging Epipolar Cues for Monocular Depth (ECCV 2026 · similarity 0.88)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.88)
- PatchFusion: An End-to-End Tile-Based Framework for High-Resolution Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.87)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.86)
- MonoInstance: Enhancing Monocular Priors via Multi-view Instance Alignment for Neural Rendering and Reconstruction (CVPR 2025 · similarity 0.86)
- DinoComplete: 3D Shape Completion with Distilled Semantic Priors and State Space Models (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
