Research map / 3D vision / NeRF and radiance fields
DyLiN: Making Light Field Networks Dynamic
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 |
|---|---|---|---|
| Distill a dynamic ray-based light field student from a pretrained dynamic NeRF, then fine-tune on real video | DyLiN · Synthetic dynamic scenes | LPIPS: 0.0663 -> 0.0184 | Helped |
| Use DyLiN instead of TiNeuVox-B | Synthetic dynamic scenes · Synthetic Lego scene | Wall-clock time (ms): 6,920.0 -> 116.0 | Helped |
| Add controllable attribute inputs and attribute-specific masked hyperspace codes | CoDyLiN · Eyes/Mouth | MS-SSIM: 0.7458 -> 0.9510 | Helped |
| Add controllable attribute inputs and attribute-specific masked hyperspace codes | CoDyLiN · Transformer | PSNR: 23.0319 -> 23.5882 | Helped |
| Remove deformation and hyperspace MLPs (ablation) | DyLiN · Synthetic dynamic scenes | PSNR: 32.07 -> 32.43 | Hurt |
| Replace explicit ray deformation and ray hyperspace MLPs with pointwise deformation MLP only (ablation) | DyLiN · Synthetic dynamic scenes | PSNR: 31.24 -> 32.43 | Hurt |
| Remove fine-tuning on real data (ablation) | DyLiN · Synthetic dynamic scenes | PSNR: 31.37 -> 32.43 | Hurt |
| Remove fine-tuning on real data (ablation) | DyLiN · Real dynamic scenes | PSNR: 24.0 -> 25.1 | Hurt |
| Use between 4 and 32 sampled points per ray rather than a larger number (ablation) | DyLiN · Standup | PSNR | No clear effect |
| Increase ray samples per ray to 64 or 128 (ablation) | DyLiN · Standup | PSNR | Hurt |
| Use smaller KD sample counts, from 100 through 5000 (ablation) | DyLiN · Standup | PSNR | Hurt |
Closest papers by meaning
- Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering (NeurIPS 2021 · similarity 0.89)
- Learning Neural Light Fields With Ray-Space Embedding (CVPR 2022 · similarity 0.89)
- BLiRF: Bandlimited Radiance Fields for Dynamic Scene Modeling (AAAI 2024 · similarity 0.88)
- Real-time Photorealistic Dynamic Scene Representation and Rendering with 4D Gaussian Splatting (ICLR 2024 · similarity 0.85)
- A General Implicit Framework for Fast NeRF Composition and Rendering (AAAI 2024 · similarity 0.85)
- NeLF-Pro: Neural Light Field Probes for Multi-Scale Novel View Synthesis (CVPR 2024 · similarity 0.85)
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.