Research map / 3D vision / Face and shape
Learning Canonical 3D Object Representation for Fine-Grained Recognition
Sunghun Joung, Seungryong Kim, Minsu Kim, Ig-Jae Kim, Kwanghoon Sohn
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 4-D positional encoding on the canonical warped feature (ablation) | Ours w/o fs · CUB-Birds | accuracy: 83.7 -> 86.8 | Improved |
| add 4-D positional encoding on the canonical warped feature (ablation) | Ours w/o fs · Stanford-Cars | accuracy: 85.5 -> 93.2 | Improved |
| add shape encoder using 3D shape deformation (ablation) | Ours w/o fs · CUB-Birds | accuracy: 86.8 -> 88.4 | Improved |
| add shape encoder using 3D shape deformation (ablation) | Ours w/o fs · Stanford-Cars | accuracy: 93.2 -> 94.7 | Improved |
| use 4-D sinusoidal coordinates instead of 2-D coordinates for positional encoding (ablation) | Ours · CUB-Birds | accuracy: 85.2 -> 86.8 | Improved |
| use 4-D sinusoidal coordinates instead of 2-D coordinates for positional encoding (ablation) | Ours · Stanford-Cars | accuracy: 89.8 -> 93.2 | Improved |
| use 3 FC layers instead of 0 in the shape encoder (ablation) | Ours · CUB-Birds | accuracy: 86.8 -> 88.4 | Improved |
| use 4 FC layers instead of 3 in the shape encoder (ablation) | Ours · CUB-Birds | accuracy: 88.4 -> 87.9 | Hurt |
| use 4 FC layers instead of 3 in the shape encoder (ablation) | Ours · Stanford-Cars | accuracy: 94.7 -> 94.4 | Hurt |
| add positional encoding to the appearance feature (ablation) | Ours w/o PE, fs · CUB-Birds 3D reconstruction | PCK: 51.5 -> 51.9 | Improved |
| add shape encoder (ablation) | Ours w/o fs · CUB-Birds 3D reconstruction | mask IoU: 0.732 -> 0.737 | Improved |
| replace 2D spatial deformation modules (VTN) with canonical 3D representation | VTN · Stanford-Cars | accuracy: 82.7 -> 85.5 | Improved |
Closest papers by meaning
- Learning Canonical View Representation for 3D Shape Recognition With Arbitrary Views (ICCV 2021 · similarity 0.87)
- Learning Category-Specific Mesh Reconstruction from Image Collections (ECCV 2018 · similarity 0.86)
- Rethinking Rotation-Invariant Recognition of Fine-Grained Shapes From the Perspective of Contour Points (TPAMI 2025 · similarity 0.86)
- Learning Canonical Shape Space for Category-Level 6D Object Pose and Size Estimation (CVPR 2020 · similarity 0.86)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · similarity 0.84)
- Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space (CVPR 2025 · similarity 0.84)
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.
