Research map / 3D vision / 3D generation
MvDeCor: Multi-View Dense Correspondence Learning for Fine-Grained 3D Segmentation
Gopal Sharma, Kangxue Yin, Subhransu Maji, Evangelos Kalogerakis, Or Litany, Sanja Fidler
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 multi-view dense correspondence pretraining vs training from scratch | 2D CNN · PartNet (k=10) | part mIoU: 32.0 -> 35.9 | Improved |
| add multi-view dense correspondence pretraining vs DenseCL | 2D CNN · PartNet (k=10) | part mIoU: 34.2 -> 35.9 | Improved |
| use ImageNet pretrained weights vs scratch | 2D CNN · PartNet (k=10) | part mIoU: 32.0 -> 32.8 | Improved |
| use 2D dense contrastive learning (DenseCL) vs ImageNet init | 2D CNN · PartNet (k=10) | part mIoU: 32.8 -> 34.2 | Improved |
| 3D scratch vs 2D scratch | segmentation network · PartNet (k=10) | part mIoU: 32.0 -> 30.3 | Hurt |
| add learned deformation-based alignment (3D) | 3D network · PartNet (k=10) | part mIoU: 30.3 -> 27.5 | Hurt |
| add MvDeCor pretraining with sparse labeled views | 2D CNN · PartNet (k=10, v=5) | part mIoU: 25.9 -> 30.3 | Improved |
| add MvDeCor pretraining on larger label set | 2D CNN · PartNet chair/table/lamp (k=30) | part mIoU: 13.7 -> 16.6 | Improved |
| add MvDeCor pretraining on RenderPeople RGB | 2D CNN · RenderPeople RGB (k=5) | part mIoU: 50.2 -> 67.5 | Improved |
| add MvDeCor pretraining with geometry input | 2D CNN · RenderPeople RGB+Geom (k=5) | part mIoU: 55.3 -> 58.8 | Improved |
| add MvDeCor pretraining with 3 labeled views and geometry | 2D CNN · RenderPeople RGB+Geom (k=5, v=3) | part mIoU: 40.6 -> 50.3 | Improved |
| use ImageNet init with depth and normal input | 2D CNN · RenderPeople RGB+Geom (k=5) | part mIoU: 55.3 -> 55.3 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports MvDeCor as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- 3x2: 3D Object Part Segmentation by 2D Semantic Correspondences (ECCV 2024 · 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
- MVTN: Learning Multi-view Transformations for 3D Understanding (IJCV 2024 · similarity 0.88)
- Learning Fine-Grained Segmentation of 3D Shapes Without Part Labels (CVPR 2021 · similarity 0.88)
- Unsupervised Learning of Dense Shape Correspondence (CVPR 2019 · similarity 0.87)
- Universal 3D Shape Matching via Coarse-to-Fine Language Guidance (CVPR 2026 · similarity 0.87)
- Learning Implicit Functions for Topology-Varying Dense 3D Shape Correspondence (NeurIPS 2020 · similarity 0.87)
- Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space (CVPR 2025 · 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.
