Research map / 3D vision / Face and shape
CoordX: Accelerating Implicit Neural Representation with a Split MLP Architecture
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Split coordinate inputs into branches and fuse their features | CoordMLPs for image fitting · DIV2K image fitting, 512×512 | PSNR: 40.25 -> 37.08; 37.30 -> 33.89 | Hurt |
| Use accelerated split-compatible sampling (ablation) | CoordX image fitting · DIV2K image fitting, 512×512 | training time: 22.33 -> 5.73 min; 27.08 -> 5.06 min | Helped |
| Split spatial coordinates from time rather than splitting X, Y, and time separately (ablation) | CoordX video fitting · Video fitting | PSNR: 31.69 -> 33.63 | Helped |
| Use accelerated split-compatible sampling (ablation) | CoordX video fitting, XY+T · Video fitting | training time: 5.22 h -> 2.13 h | Helped |
| Split coordinates into branches | CoordX 3D shape fitting · 3D shape fitting, easy and hard | IoU: 0.991 / 0.970 -> 0.991 / 0.970 | No clear effect |
| Use four-branch CoordX in place of the LIIF decoder MLP | LIIF with EDSR · DIV2K super-resolution, ×2 | PSNR: 34.67 -> 34.65 | No clear effect |
| Use four-branch CoordX in place of the LIIF decoder MLP | LIIF with EDSR · DIV2K super-resolution, ×2 inference | inference time: 131.03 ms -> 45.53 ms | Helped |
| Increase the number of split layers by reducing D_f (ablation) | CoordX 3D shape fitting · 3D shape fitting, 128³ grid | IoU: 0.976 -> 0.880 | Hurt |
| Increase the number of split layers by reducing D_f (ablation) | CoordX 3D shape fitting · 3D shape fitting, 128³ grid | training time: 5.72 -> 2.38 min | Helped |
| Increase the reduction dimension R (ablation) | CoordX image fitting · DIV2K image fitting | PSNR: 37.08 -> 39.70; 37.08 -> 41.60 | Helped |
| At fixed R, widen splitting layers instead of only widening the layer before fusion (ablation) | CoordX image fitting · DIV2K image fitting | PSNR: 39.70 -> 41.32; 41.60 -> 43.55 | Helped |
| Replace multiplicative feature fusion with additive fusion (ablation) | CoordX image fitting · Image fitting, SIREN and positional encoding | PSNR: 36.32 -> 33.03; 31.41 -> 26.51 | Hurt |
Closest papers by meaning
- PINs: Progressive Implicit Networks for Multi-Scale Neural Representations (ICML 2022 · similarity 0.87)
- Implicit Zoo: A Large-Scale Dataset of Neural Implicit Functions for 2D Images and 3D Scenes (NeurIPS 2024 · similarity 0.86)
- Coordinate-Aware Modulation for Neural Fields (ICLR 2024 · similarity 0.85)
- MeshXL: Neural Coordinate Field for Generative 3D Foundation Models (NeurIPS 2024 · similarity 0.85)
- CoFie: Learning Compact Neural Surface Representations with Coordinate Fields (NeurIPS 2024 · similarity 0.85)
- Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding (AAAI 2024 · 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.