Research map / 3D vision / Gaussian splatting
TrackNeRF: Bundle Adjusting NeRF from Sparse and Noisy Views via Feature Tracks
Jinjie Mai, Wenxuan Zhu, Sara Rojas, Jesus Zarzar, Abdullah Hamdi, Guocheng Qian, Bing Li, Silvio Giancola, Bernard Ghanem
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 |
|---|---|---|---|
| add feature-track reprojection loss to BARF (ablation) | BARF · DTU (3 views, 15% noise) | PSNR: 10.70 -> 18.07 | Improved |
| add track keypoint refinement (ablation) | TrackNeRF · DTU (3 views, 15% noise) | PSNR: 18.07 -> 18.39 | Improved |
| add depth regularization loss (ablation) | TrackNeRF · DTU (3 views, 15% noise) | PSNR: 18.39 -> 18.53 | Improved |
| add feature-track loss on forward-facing LLFF | TrackNeRF vs SPARF · LLFF (3 views, no poses) | PSNR: 19.58 -> 19.60 | No clear effect |
| allow longer feature tracks (max length 2 to 6) (ablation) | TrackNeRF · DTU 6 views (one scene) | PSNR: 13.15 -> 19.56 | Improved |
| increase initial pose noise from 15% to 45% (ablation) | TrackNeRF · DTU 3 views (one scene) | PSNR: 16.14 -> 8.34 | Hurt |
| add track keypoint adjustment to SPARF (ablation) | SPARF · DTU (3 views, 15% noise) | PSNR: 17.74 -> 17.84 | Improved |
| apply depth loss in both stages instead of only stage 2 (ablation) | TrackNeRF · DTU (3 views, 15% noise) | Rotation error (deg): 1.12 -> 1.49 | Hurt |
| increase depth regularization weight to 10^0 or lower to 10^-2 (ablation) | TrackNeRF · DTU | visual and numeric quality | Hurt |
| use global feature tracks instead of pairwise (SPARF) correspondences, 6 views | TrackNeRF vs SPARF · DTU (6 noisy views) | PSNR: 21.39 -> 22.78 | Improved |
| reduce training to 23K iterations (ablation) | TrackNeRF · DTU (3 views, 15% noise) | PSNR (masked): 19.65 -> 19.35 at 10.3h vs 27.4h | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- SPARF — SPARF: Neural Radiance Fields From Sparse and Noisy Poses (CVPR 2023) · better on 9 of 11 shared results
- ReconFusion — ReconFusion: 3D Reconstruction with Diffusion Priors (CVPR 2024) · better on 1 of 1 shared result
- BARF — BARF: Bundle-Adjusting Neural Radiance Fields (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
- ProvNeRF: Modeling per Point Provenance in NeRFs as a Stochastic Field (NeurIPS 2024 · similarity 0.87)
- Joint Optimization of Neural Radiance Fields and Continuous Camera Motion from a Monocular Video (CVPR 2025 · similarity 0.87)
- STGC-NeRF: Spatial-Temporal Geometric Consistency for LiDAR Neural Radiance Fields in Dynamic Scenes (AAAI 2025 · similarity 0.87)
- ColNeRF: Collaboration for Generalizable Sparse Input Neural Radiance Field (AAAI 2024 · similarity 0.86)
- ICON: Incremental CONfidence for Joint Pose and Radiance Field Optimization (CVPR 2024 · similarity 0.86)
- AltNeRF: Learning Robust Neural Radiance Field via Alternating Depth-Pose Optimization (AAAI 2024 · similarity 0.86)
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.
