Research map / 3D vision / Face and shape
LiTo: Surface Light Field Tokenization
Jen-Hao Rick Chang, Xiaoming Zhao, Dorian Chan, Oncel Tuzel
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 |
|---|---|---|---|
| encode surface light field with view-dependent Gaussians instead of TRELLIS view-independent latent | LiTo · Toys4k reconstruction Hard cameras | PSNR: 27.57 -> 32.36 | Improved |
| increase spherical harmonics degree from 0 to 3 (ablation) | LiTo · Toys4k Hard (TRELLIS lighting) | PSNR: 30.21 -> 32.36 | Improved |
| add viewing direction (ray) to encoder with SH degree 0 (ablation) | LiTo · Toys4k Hard (TRELLIS lighting) | LPIPS: 0.067 -> 0.068 | No clear effect |
| add viewing direction (ray) to encoder with SH degree 3 (ablation) | LiTo · Toys4k Hard (TRELLIS lighting) | LPIPS: 0.058 -> 0.056 | Improved |
| add viewing direction (ray) to encoder with SH degree 3 (ablation) | LiTo · Toys4k Hard (TRELLIS lighting) | PSNR: 32.11 -> 32.36; 35.54 -> 35.32 | Mixed |
| add ray (camera intrinsics) conditioning to the image-to-3D generator (ablation) | LiTo generator · Toys4k generation | overall performance | Hurt |
| use LiTo instead of TRELLIS for generation, conditioning view | LiTo · Toys4k conditioning view | FID: 12.84 -> 6.219 | Improved |
| use LiTo instead of TRELLIS for generation, novel views | LiTo · Toys4k novel view | FID: 7.600 -> 6.216 | No clear effect |
| use LiTo instead of TRELLIS for generation, novel views | LiTo · Toys4k novel view | KID_dino: 3.166 -> 3.522 | Hurt |
| use LiTo latent (no coarse geometry oracle) vs Shape Tokens | LiTo · Toys4k geometry | Chamfer distance: 119.8 -> 88.30 | Improved |
| use LiTo without oracle vs TripoSF with oracle | LiTo (mesh decoder) · Toys4k geometry | Chamfer distance: 76.99 -> 80.55 | Hurt |
| add mesh decoder (ablation) | LiTo · Toys4k geometry | Chamfer distance: 88.30 -> 80.55 | Improved |
Closest papers by meaning
- Light Field Neural Rendering (CVPR 2022 · similarity 0.90)
- Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering (NeurIPS 2021 · similarity 0.89)
- Multiview Neural Surface Reconstruction by Disentangling Geometry and Appearance (NeurIPS 2020 · similarity 0.88)
- Latent Radiance Fields with 3D-aware 2D Representations (ICLR 2025 · similarity 0.88)
- GRF: Learning a General Radiance Field for 3D Representation and Rendering (ICCV 2021 · similarity 0.88)
- Generative Multiview Relighting for 3D Reconstruction under Extreme Illumination Variation (CVPR 2025 · similarity 0.88)
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.
