Research map / 3D vision / 3D generation
Pano3D: Unified 3D Reconstruction and Panoptic Segmentation
Victor Barberteguy, Ahmet Iscen, Mathilde Caron, Alireza Fathi, Gül Varol, Cordelia Schmid
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 |
|---|---|---|---|
| joint finetuning of geometry decoder and segmentation head vs frozen backbone (ablation) | Pano3D (MUSt3R) · ScanNet val | AP3D: 10.8 -> 15.7 | Improved |
| joint finetuning vs frozen backbone (ablation) | Pano3D (MUSt3R) · ScanNet val | 3D mIoU: 61.9 -> 65.3 | Improved |
| finetune semantics only without geometry loss (ablation) | Pano3D (MUSt3R) · ScanNet val | AbsRel (MVD): 3.5 -> 156.3 | Hurt |
| joint training vs frozen MUSt3R for multi-view depth (ablation) | Pano3D (MUSt3R) · ScanNet val | AbsRel (MVD): 5.2 -> 3.5 | Improved |
| unfreeze encoder for end-to-end training (ablation) | Pano3D (MUSt3R) · ScanNet val | 3D mIoU: 65.3 -> 63.4 mIoU3D; AP 15.7 -> 16.6; AbsRel 3.5 -> 3.3 | Mixed |
| combine encoder and decoder features (ablation) | Pano3D (MUSt3R) · ScanNet val | 3D mIoU: 60.4 -> 65.3 | Improved |
| use decoder features only (no encoder skip) (ablation) | Pano3D (MUSt3R) · ScanNet val | 3D mIoU: 65.3 -> 63.8 | Hurt |
| use equal geometry and semantic loss weights (ablation) | Pano3D (MUSt3R) · ScanNet val | AbsRel (MVD): 3.5 -> 5.1 | Hurt |
| raise geometry weight to 20x semantic (ablation) | Pano3D (MUSt3R) · ScanNet val | 3D mIoU: 65.3 -> 63.3 | Mixed |
| scale segmentation gradients into geometry decoder by 0.1 (ablation) | Pano3D (Pi3) · ScanNet++ | MVD AbsRel: 2.95 -> 2.77 | Improved |
| scale segmentation gradients by 0.1 (ablation) | Pano3D (Pi3) · ScanNet++ | mIoU: 47.41 -> 44.91 | Hurt |
| add DINOv2 features to mask decoder (ablation) | Pano3D · ScanNet val | mIoU2D: 73.5 -> 71.7 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PanSt3R — PanSt3R: Multi-view Consistent Panoptic Segmentation (ICCV 2025) · better on 3 of 3 shared results
- IGGT — IGGT: Instance-Grounded Geometry Transformer for Semantic 3D Reconstruction (ICLR 2026) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PanoRecon: Real-Time Panoptic 3D Reconstruction from Monocular Video (CVPR 2024 · similarity 0.92)
- Uni-3D: A Universal Model for Panoptic 3D Scene Reconstruction (ICCV 2023 · similarity 0.91)
- Panoptic 3D Scene Reconstruction From a Single RGB Image (NeurIPS 2021 · similarity 0.91)
- Multi-view Consistent 3D Panoptic Scene Understanding (AAAI 2025 · similarity 0.87)
- PanoOcc: Unified Occupancy Representation for Camera-based 3D Panoptic Segmentation (CVPR 2024 · similarity 0.87)
- UniSDF: Unifying Neural Representations for High-Fidelity 3D Reconstruction of Complex Scenes with Reflections (NeurIPS 2024 · 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.
