Research map / 3D vision / NeRF and radiance fields
NeXT: Towards High Quality Neural Radiance Fields via Multi-Skip Transformer
Yunxiao Wang, Yanjie Li, Peidong Liu, Tao Dai, Shu-Tao Xia
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 |
|---|---|---|---|
| replace NeRF MLP with ray-level NeXT-S transformer at similar parameters | NeRF · Blender | PSNR: 31.01 -> 32.34 | Improved |
| replace MLP with vanilla Transformer without multi-skip | NeRF · Blender | PSNR: 31.66 -> 31.44 | Hurt |
| scale NeXT from S to L | NeXT · Blender | PSNR: 32.34 -> 34.04 | Improved |
| make NeRF MLP deeper (20 layers) | NeRF · Blender | PSNR | Hurt |
| widen and deepen NeRF MLP to D16-W384 | NeRF · Blender | PSNR | No clear effect |
| increase sample points from 192 to 256 | NeXT-L · Blender | PSNR: +0.85 PSNR | Improved |
| add 1D sinusoidal positional embedding (ablation) | NeXT-B · Blender | PSNR: 32.19 -> 32.72 | Improved |
| use learnable instead of sinusoidal positional embedding (ablation) | NeXT-B · Blender | PSNR: 32.72 -> 32.44 | Hurt |
| use local-window self-attention (window 64) instead of global attention (ablation) | NeXT-B · Blender | PSNR: 32.25 -> 32.34 | No clear effect |
| shrink local window from 64 to 32 (ablation) | NeXT-B · Blender | PSNR: 32.34 -> 32.26 | No clear effect |
| add multi-skip connection (ablation) | NeXT · Blender | PSNR: 31.44 -> 32.44 | Improved |
| add multi-skip connection at layers 2,4,6 (ablation) | NeRF (JAX) · Blender | PSNR: 31.66 -> 31.90 | 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.
- Mip-NeRF — Mip-NeRF: A Multiscale Representation for Anti-Aliasing Neural Radiance Fields (ICCV 2021) · better on 4 of 4 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
- Efficient View Synthesis with Neural Radiance Distribution Field (ICCV 2023 · similarity 0.92)
- Cross-Ray Neural Radiance Fields for Novel-View Synthesis from Unconstrained Image Collections (ICCV 2023 · similarity 0.91)
- Neural Radiance Fields From Sparse RGB-D Images for High-Quality View Synthesis (TPAMI 2022 · similarity 0.91)
- Rad-NeRF: Ray-decoupled Training of Neural Radiance Field (NeurIPS 2024 · similarity 0.91)
- R2L: Distilling Neural Radiance Field to Neural Light Field for Efficient Novel View Synthesis (ECCV 2022 · similarity 0.91)
- NeRF-VPT: Learning Novel View Representations with Neural Radiance Fields via View Prompt Tuning (AAAI 2024 · similarity 0.90)
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.
