Research map / 3D vision / 3D generation
ViewFormer: NeRF-Free Neural Rendering from Few Images Using Transformers
Jonáš Kulhánek, Erik Derner, Torsten Sattler, Robert Babuška
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 |
|---|---|---|---|
| remove the localization (pose estimation) head, i.e. train for view synthesis only (ablation) | ViewFormer @ all cat. · CO3D (all categories, avg test) | PSNR: 15.3 -> 15.4 | No clear effect |
| train on all CO3D categories instead of only 10 categories (ablation) | ViewFormer · CO3D (10 categories, avg test) | PSNR: 15.6 -> 16.0 | Improved |
| increase number of context views from 1 to 9 (ablation) | ViewFormer @ all cat. · CO3D (all categories, test) | PSNR: 13.7 -> 16.1 | Improved |
| remove the perceptual loss from the codebook training (ablation) | VQ-VAE codebook · codebook reconstruction | image quality (qualitative) | Hurt |
| use dynamic loss weighting for position and orientation instead of plain sum (ablation) | ViewFormer localization · pose loss | pose accuracy | Hurt |
| replace explicit-3D NeRF-based NerFormer by 2D-only ViewFormer (training cost) | CO3D all categories · CO3D | training GPU-hours: 8400 -> 850 GPU-hours | Improved |
| replace NerFormer by ViewFormer (render time per view) | CO3D · CO3D | render time: 178 s -> 93 ms | Improved |
| ViewFormer vs NerFormer on LPIPS (all categories) | ViewFormer · CO3D (all categories, avg test) | LPIPS: 0.24 -> 0.23 | Improved |
| ViewFormer vs NerFormer on PSNR (10 categories) | ViewFormer @ all cat. · CO3D (10 categories, avg test) | PSNR: 17.6 -> 16.0 | Hurt |
| ViewFormer vs GQN-style baselines on SM7 | ViewFormer · Shepard-Metzler-7-Parts | MAE: 2.14 -> 1.61 | Improved |
Later papers that compared with this method
7 later accepted papers report ViewFormer as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- Streaming Dense Voxel Representations for 3D Occupancy Prediction (ECCV 2026 · better on 2 of 2 shared results)
- Dr.Occ: Depth- and Region-Guided 3D Occupancy from Surround-View Cameras for Autonomous Driving (CVPR 2026 · better on 1 of 1 shared result)
- DreamComposer++: Empowering Diffusion Models with Multi-View Conditions for 3D Content Generation (TPAMI 2025 · better on 3 of 3 shared results)
- Occupancy Learning with Spatiotemporal Memory (ICCV 2025 · better on 1 of 1 shared result)
- GSD: View-Guided Gaussian Splatting Diffusion for 3D Reconstruction (ECCV 2024 · better on 3 of 3 shared results)
- SparseGNV: Generating Novel Views of Indoor Scenes with Sparse RGB-D Images (AAAI 2024 · better on 2 of 2 shared results)
- Generative Novel View Synthesis with 3D-Aware Diffusion Models (ICCV 2023 · better on 2 of 2 shared results)
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.
- STR-GQN — STR-GQN: Scene Representation and Rendering for Unknown Cameras Based on Spatial Transformation Routing (ICCV 2021) · better on 1 of 1 shared result
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
- Putting NeRF on a Diet: Semantically Consistent Few-Shot View Synthesis (ICCV 2021 · similarity 0.92)
- GoLF-NRT: Integrating Global Context and Local Geometry for Few-Shot View Synthesis (CVPR 2025 · similarity 0.90)
- CodecNeRF: Toward Fast Encoding and Decoding, Compact, and High-quality Novel-view Synthesis (AAAI 2025 · similarity 0.90)
- NeRF-VPT: Learning Novel View Representations with Neural Radiance Fields via View Prompt Tuning (AAAI 2024 · similarity 0.89)
- NeRF Director: Revisiting View Selection in Neural Volume Rendering (CVPR 2024 · similarity 0.89)
- Few-Shot Neural Radiance Fields under Unconstrained Illumination (AAAI 2024 · similarity 0.89)
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.
