Research map / 3D vision / Depth and stereo
FrozenRecon: Pose-free 3D Scene Reconstruction with Frozen Depth Models
Guangkai Xu, Wei Yin, Hao Chen, Chunhua Shen, Kai Cheng, Feng Zhao
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 geometric consistency alignment module instead of directly optimizing per-frame depth (ablation) | FrozenRecon · NYU | F-score: 0.303 -> 0.622 | Improved |
| use global plus local scale-shift alignment instead of RCVD deformation map (ablation) | FrozenRecon · NYU | AbsRel: 0.153 -> 0.092 | Improved |
| add photometric constraint (ablation) | FrozenRecon · ScanNet | F-score: 0.169 -> 0.410 | Improved |
| add geometric constraint (ablation) | FrozenRecon · ScanNet | F-score: 0.323 -> 0.410 | Improved |
| add regularization constraint on anchor point weights (ablation) | FrozenRecon · ScanNet | F-score: 0.254 -> 0.410 | Improved |
| add global keyframe stage after local stage (ablation) | FrozenRecon · NYU | AbsRel: 0.111 -> 0.092 | Improved |
| add global keyframe stage after local stage (ablation) | FrozenRecon · NYU | Chamfer C-l1: 0.091 -> 0.099 | Hurt |
| replace photometric constraint with optical-flow constraint (ablation) | FrozenRecon · NYU | F-score: 0.622 vs 0.627 | No clear effect |
| use pose-based keyframe sampling instead of flow-guided sampling (ablation) | FrozenRecon · NYU | F-score: 0.565 -> 0.622 | Improved |
| provide GT intrinsics only (ablation) | FrozenRecon · NYU | F-score: 0.622 -> 0.622 | No clear effect |
| provide GT poses (ablation) | FrozenRecon · NYU | F-score: 0.622 -> 0.662 | Improved |
| FrozenRecon vs baselines on outdoor KITTI depth | FrozenRecon · KITTI | AbsRel: 0.203 vs 0.112 (BoostingDepth-DROID) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- DROID-SLAM — DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras (NeurIPS 2021) · better on 3 of 5 shared results
- DPSNet* — DPSNet: End-to-end Deep Plane Sweep Stereo (ICLR 2019) · better on 0 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
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.88)
- LEAP: Liberate Sparse-View 3D Modeling from Camera Poses (ICLR 2024 · similarity 0.87)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.87)
- Zero-Shot Scene Reconstruction from Single Images with Deep Prior Assembly (NeurIPS 2024 · similarity 0.87)
- Know Your Neighbors: Improving Single-View Reconstruction via Spatial Vision-Language Reasoning (CVPR 2024 · similarity 0.87)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.87)
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.
