Research map / 3D vision / Depth and stereo
BA-Net: Dense Bundle Adjustment Networks
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 |
|---|---|---|---|
| learn features end-to-end through feature-metric bundle adjustment instead of using ImageNet-pretrained features (ablation) | BA-Net · ScanNet test | RMSE (linear): 0.481 -> 0.346 | Helped |
| jointly optimize depth and camera poses with bundle adjustment rather than estimate pose with fixed depth (ablation) | BA-Net · ScanNet test | RMSE (linear): 0.428 -> 0.346 | Helped |
| use learned LM damping instead of removing lambda (ablation) | BA-Net · ScanNet test | RMSE (linear): 0.763 -> 0.346 | Helped |
| adapt lambda with an MLP instead of using a constant lambda (ablation) | BA-Net · ScanNet | camera pose error | Helped |
| increase the number of reconstructed views from 2 to 5 | BA-Net · ScanNet multi-view | RMSE (linear): 0.346 -> 0.223 | Helped |
| use BA-Net instead of DeMoN | two-view SfM · ScanNet test | RMSE (linear): 0.761 -> 0.346 | Helped |
| use end-to-end feature-metric BA with learned basis depths instead of CodeSLAM's separately learned depth parameterization | SfM on EuroC MH02 · EuroC MH02 | translation error: roughly 1 m -> about 0.4 m | Helped |
| use BA-Net instead of DeMoN | two-view SfM · Scenes11 | L1-inv: 0.019 -> 0.08 | Hurt |
Closest papers by meaning
- Learning Structure-from-Motion with Graph Attention Networks (CVPR 2024 · similarity 0.87)
- BA-T: An Iterative Transformer for Two-View Bundle Adjustment (NeurIPS 2026 · similarity 0.85)
- Dense-SfM: Structure from Motion with Dense Consistent Matching (CVPR 2025 · similarity 0.84)
- VGGSfM: Visual Geometry Grounded Deep Structure From Motion (CVPR 2024 · similarity 0.84)
- SuperPrimitive: Scene Reconstruction at a Primitive Level (CVPR 2024 · similarity 0.83)
- Light3R-SfM: Towards Feed-forward Structure-from-Motion (CVPR 2025 · similarity 0.83)
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.