Research map / 3D vision / Face and shape
Modulated Periodic Activations for Generalizable Local Functional Representations
Ishit Mehta, Michaël Gharbi, Connelly Barnes, Eli Shechtman, Ravi Ramamoorthi, Manmohan Chandraker
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 |
|---|---|---|---|
| use modulation network instead of concatenating latent code with coordinates in a sine MLP (ablation) | sine synthesis network · CelebA 32x32 | PSNR: 20.15 -> 29.42 | Improved |
| use modulation network instead of concatenation on a harder dataset (ablation) | sine synthesis network · CIFAR-10 | PSNR: 25.73 | Improved |
| use modulator instead of hypernetwork conditioning (ablation) | SIREN synthesis network · CelebA 32x32 | PSNR: 26.70 -> 29.42 | Improved |
| use sine activations instead of ReLU in the synthesis network (local model) (ablation) | local functional representation · Div2K 2K images | PSNR: 34.73 -> 38.03 | Improved |
| use sine activations instead of ReLU in the synthesis network (local model) (ablation) | local functional representation · ThreedScans Scene A | Chamfer distance (x1e-5): 1.37 -> 1.32 | Improved |
| use sine activations instead of ReLU in the synthesis network (local model) (ablation) | local functional representation · ThreedScans Scene B | Chamfer distance (x1e-5): 8.37 -> 2.40 | Improved |
| add local tiling with pretrained generalizable model instead of per-image global overfitting | high-resolution image encoding · Div2K 2K images | PSNR: 28.48 -> 38.03 | Improved |
| add local tiling with pretrained model to video encoding | high-res video encoding · 6 pexels videos 640x256 | PSNR: 25.19 -> 25.21 | No clear effect |
| use FFN sigma=10 instead of sigma=1 (ablation) | FFN conditional baseline · CelebA 32x32 (2x resolution) | PSNR: 24.37 -> 4.22 | Hurt |
| use larger convolutional encoder (17M params) (ablation) | Ours · CelebA 32x32 | PSNR: 29.42 -> 29.64 | Improved |
| use modulator network instead of DeepSDF ReLU concatenation | conditional SDF network · ShapeNet cars | Chamfer distance (median): 0.00284 -> 0.00230 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepSDF — DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation (CVPR 2019) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Neural Functions for Learning Periodic Signal (ICLR 2025 · similarity 0.87)
- A sampling theory perspective on activations for implicit neural representations (ICML 2024 · similarity 0.86)
- FINER: Flexible Spectral-bias Tuning in Implicit NEural Representation by Variable-periodic Activation Functions (CVPR 2024 · similarity 0.86)
- Coordinate-Aware Modulation for Neural Fields (ICLR 2024 · similarity 0.85)
- Few-shot Implicit Function Generation via Equivariance (CVPR 2025 · similarity 0.84)
- FreSh: Frequency Shifting for Accelerated Neural Representation Learning (ICLR 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.
