Research map / 3D vision / Face and shape
Pixel2Mesh++: 3D Mesh Generation and Refinement From Multi-View Images
Chao Wen, Yinda Zhang, Chenjie Cao, Zhuwen Li, Xiangyang Xue, Yanwei Fu
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 Deformation Network refinement on top of MVP2M coarse mesh | MVP2M · ShapeNet multi-view (mean over 13 categories) | F-score(tau): 61.20 -> 67.23 | Improved |
| Add MDN refinement on top of MVDISN coarse mesh | MVDISN · ShapeNet multi-view (mean over 13 categories) | F-score(tau): 70.78 -> 75.24 | Improved |
| Add MDN refinement on top of MVP2M | MVP2M · ShapeNet multi-view (mean over 13 categories) | Chamfer Distance: 0.456 -> 0.381 | Improved |
| Add MDN refinement on top of MVDISN | MVDISN · ShapeNet multi-view (mean over 13 categories) | volumetric IoU: 0.527 -> 0.508 | Hurt |
| Add MDN refinement on top of MVP2M | MVP2M · ShapeNet multi-view (mean over 13 categories) | volumetric IoU: 0.411 -> 0.436 | Improved |
| Add multi-view feature pooling to Pixel2Mesh (MVP2M vs P2M-M) | P2M-M · ShapeNet multi-view (mean over 13 categories) | F-score(tau): 61.72 -> 61.20 | No clear effect |
| Add feature statistics (mean, max, std) instead of concatenation (ablation) | Ours-P · ShapeNet ablation (Ours-P, known camera) | F-score(tau): 65.26 -> 67.32 | Improved |
| Add feature statistics instead of concatenation (ablation) | Ours-P · ShapeNet ablation (Ours-P, known camera) | Chamfer Distance: 0.511 -> 0.381 | Improved |
| Add re-sampled Chamfer loss instead of vertex-only Chamfer (ablation) | Ours-P · ShapeNet ablation (Ours-P, known camera) | F-score(tau): 66.26 -> 67.32 | Improved |
| Add re-sampled Chamfer loss (ablation) | Ours-P · ShapeNet ablation (Ours-P, known camera) | Chamfer Distance: 0.496 -> 0.381 | Improved |
| Increase test views from 2 to 5 (trained on 3 views) | Ours-P (MDN) · ShapeNet views | F-score(tau): 64.48 -> 68.29 | Improved |
| Use predicted camera pose instead of ground-truth | Ours-P · ShapeNet robustness | F-score(tau): 67.32 -> 63.22 | Hurt |
Closest papers by meaning
- Pixel2Mesh++: Multi-View 3D Mesh Generation via Deformation (ICCV 2019 · similarity 0.99)
- Pixel2Mesh: 3D Mesh Model Generation via Image Guided Deformation (TPAMI 2020 · similarity 0.94)
- Geometry-Aware Network for Non-Rigid Shape Prediction From a Single View (CVPR 2018 · similarity 0.90)
- Synthesizing 3D Shapes via Modeling Multi-View Depth Maps and Silhouettes With Deep Generative Networks (CVPR 2017 · similarity 0.90)
- Learning Generative Models of Textured 3D Meshes From Real-World Images (ICCV 2021 · similarity 0.90)
- Visual Object Networks: Image Generation with Disentangled 3D Representations (NeurIPS 2018 · similarity 0.89)
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.
