Research map / 3D vision / Depth and stereo
Mirror3D: Depth Refinement for Mirror Surfaces
Jiaqi Tan, Weijie Lin, Angel X. Chang, Manolis Savva
Check 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 |
|---|---|---|---|
| Apply Mirror3DNet to raw sensor depth | NYUv2-ref sensor depth · NYUv2-ref test, mirror pixels | RMSE: 1.272 -> 0.766 | Improved |
| Apply Mirror3DNet after raw-trained saic | saic · NYUv2-ref test, mirror pixels | RMSE: 1.245 -> 0.749 | Improved |
| Apply Mirror3DNet after raw-trained BTS | BTS · NYUv2-ref test, mirror pixels | RMSE: 0.903 -> 0.567 | Improved |
| Apply Mirror3DNet after raw-trained VNL | VNL · NYUv2-ref test, mirror pixels | RMSE: 4.216 -> 3.245 | Improved |
| Apply Mirror3DNet after raw-trained saic | saic · NYUv2-raw test, mirror pixels | RMSE: 0.107 -> 0.825 | Hurt |
| Use 3 rather than 10 anchor normals (ablation) | Mirror3DNet · Matterport3D val | Seg-AP: 0.464 -> 0.396 | Hurt |
| Use 5 rather than 10 anchor normals (ablation) | Mirror3DNet · Matterport3D val | Seg-AP: 0.464 -> 0.419 | Hurt |
| Use 7 rather than 10 anchor normals (ablation) | Mirror3DNet · Matterport3D val | Seg-AP: 0.464 -> 0.434 | Hurt |
| Use 12 rather than 10 anchor normals (ablation) | Mirror3DNet · Matterport3D val | Seg-AP: 0.464 -> 0.386 | Hurt |
| Use a 15-pixel rather than 25-pixel mirror border (ablation) | Mirror3DNet · MP3D-mesh-ref | RMSE: 0.839 -> 0.851 | Hurt |
| Use a 35-pixel rather than 25-pixel mirror border (ablation) | Mirror3DNet · MP3D-mesh-ref | RMSE: 0.839 -> 0.842 | Hurt |
Closest papers by meaning
- Depth-Aware Mirror Segmentation (CVPR 2021 · similarity 0.89)
- Learning Depth Estimation for Transparent and Mirror Surfaces (ICCV 2023 · similarity 0.89)
- Seeing Beyond Illusion: Generalized and Efficient Mirror Detection (AAAI 2026 · similarity 0.86)
- MIRROR: Make Your Object-Level Multi-View Generation More Consistent with Training-Free Rectification (ICML 2025 · similarity 0.83)
- NeRSP: Neural 3D Reconstruction for Reflective Objects with Sparse Polarized Images (CVPR 2024 · similarity 0.83)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.83)
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.
