Research map / 3D vision / Face and shape
3D VR Sketch Guided 3D Shape Prototyping and Exploration
Ling Luo, Pinaki Nath Chowdhury, Tao Xiang, Yi-Zhe Song, Yulia Gryaditskaya
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 |
|---|---|---|---|
| train encoder jointly on sketches and shapes instead of sketches only or shapes only (ablation) | sketch-to-shape auto-encoder · VR chair sketches (ShapeNet, test 202) | F_shape(G^{e^c}): 0.834 -> 0.357 (vs Shape AE); 0.437 -> 0.357 (vs Sketch AE L_SDF) | Improved |
| add L_SDF to L_L1 losses (ablation) | joint auto-encoder · VR chair sketches (ShapeNet, test 202) | F_shape(G^{e^g}): 0.199 -> 0.140 | Improved |
| add NCE contrastive losses (ablation) | joint auto-encoder · ShapeNet chair shapes (shape input) | F_shape(G^{e^g}): 0.140 -> 0.126 | Improved |
| add sketch fidelity loss L_sketch to the encoder losses (ablation) | joint auto-encoder · VR chair sketches (ShapeNet, test 202) | F_shape(G^{e^c}): 0.373 -> 0.357 | Improved |
| train CNF on inverted auto-decoder codes d^g instead of encoder codes e^g (ablation) | CNF · VR chair sketches, CNF generation | diversity D_gnrtns: 0.043 -> 0.165 | Improved |
| train CNF on inverted codes d^g instead of e^g (ablation) | CNF · VR chair sketches, CNF generation | shape fidelity to reference F_shape^avg: 0.380 -> 0.420 | Hurt |
| add sketch consistency loss L_sketch,CNF to CNF training (K=1) (ablation) | CNF(d^g) · VR chair sketches, CNF generation | sketch fidelity F_sketch^avg: 0.030 -> 0.019 | Improved |
| increase K samples in L_sketch,CNF from 1 to 8 (ablation) | CNF(d^g) + L_sketch,CNF · VR chair sketches, CNF generation | sketch fidelity F_sketch^avg: 0.019 -> 0.017 | Improved |
| K=4 instead of K=1 in L_sketch,CNF (ablation) | CNF(d^g) + L_sketch,CNF · VR chair sketches, CNF generation | F_shape(G^{e^c}): 0.366 -> 0.397 | Hurt |
| use 3D VR sketches instead of 2D sketches as input | sketch encoder with L_align and L_SDF · VR chair sketches (ShapeNet, test 202) | F_shape: 0.833 -> 0.437 | Improved |
| use clean curve networks instead of freehand sketches as input | CNF generation · ModelNet chairs (139) | F_shape(G^{e^c}) and diversity: 0.366 -> 0.237 (shape fidelity); diversity 0.161 -> 0.142 | Improved |
| generate with CNF instead of retrieving top-5 shapes | sketch-to-shape pipeline · VR chair sketches, retrieval comparison | F_shape^avg: 1.132 vs 0.431 | Improved |
Closest papers by meaning
- Order Matters: 3D Shape Generation from Sequential VR Sketches (CVPR 2026 · similarity 0.91)
- Rapid 3D Model Generation with Intuitive 3D Input (CVPR 2024 · similarity 0.91)
- Doodle Your 3D: From Abstract Freehand Sketches to Precise 3D Shapes (CVPR 2024 · similarity 0.87)
- Sketch2Mesh: Reconstructing and Editing 3D Shapes From Sketches (ICCV 2021 · similarity 0.87)
- Sketch2Model: View-Aware 3D Modeling From Single Free-Hand Sketches (CVPR 2021 · similarity 0.87)
- Diff3DS: Generating View-Consistent 3D Sketch via Differentiable Curve Rendering (ICLR 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.
