Research map / 3D vision / Depth and stereo
ULTRA-360: Unconstrained Dataset for Large-scale Temporal 3D Reconstruction across Altitudes and Omnidirectional Views
Xijun Liu, Zhaoliang Zhang, Yuxiang Guo, Yifan Zhou, Rama Chellappa, Cheng Peng
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 SuperPoint or DaD keypoint filtering to RoMa matches (ablation) | RoMa · ULTRA-360 cross-elevation matching | AUC@10: 0.0854 -> 0.6941 (Building #10) | Improved |
| use classic or other matchers (SIFT, SP+SG, LoFTR) instead of RoMa-based ones (ablation) | cross-elevation calibration · ULTRA-360 cross-elevation matching | AUC@10: 0 AUC@10 on all seven buildings | Hurt |
| train on ground+aerial images instead of single elevation (ablation) | Octree-GS, Scaffold-GS, Block-MERF · ULTRA-360 aerial test | PSNR: Block-MERF 27.451 -> 13.453; Octree-GS 29.950 -> 26.488 | Hurt |
| train on ground+aerial instead of ground only (ablation) | Octree-GS · ULTRA-360 ground test | PSNR: 21.360 -> 21.184 | No clear effect |
| combine ground and aerial training data (ablation) | Gaussian splatting methods · ULTRA-360 | number of Gaussians: EVER 535701 (G) -> 262366 (GA) | Hurt |
| add implicit neural sky model (ablation) | Octree-GS · ULTRA-360 aerial renderings from ground-only reconstruction | floaters (qualitative) | Improved |
| use farthest training image embedding instead of test image embedding (ablation) | Wild-GS · ULTRA-360 multi-appearance | PSNR: 28.133 -> 22.506 | Hurt |
| use farthest training image embedding instead of test image embedding (ablation) | Gaussian-Wild · ULTRA-360 multi-appearance | PSNR: 26.528 -> 25.621 | Hurt |
| replace per-image embedding with time-based embedding (ablation) | Wild-GS · ULTRA-360 multi-appearance | PSNR: 28.133 -> 27.973 | No clear effect |
| add rotation points to Procrustes alignment (ablation) | Procrustes Alignment · Mip-NeRF 360 sparse alignment | rotation error ER: 0.196 -> 0.156 | Improved |
| add RANSAC to Procrustes alignment (ablation) | Procrustes Alignment · Mip-NeRF 360 sparse alignment | rotation error ER: 0.196 -> 0.179 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Wild-GS — Wild-GS: Real-Time Novel View Synthesis from Unconstrained Photo Collections (NeurIPS 2024) · 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
- RPG360: Robust 360 Depth Estimation with Perspective Foundation Models and Graph Optimization (NeurIPS 2025 · similarity 0.86)
- Extreme Rotation Estimation in the Wild (CVPR 2025 · similarity 0.86)
- SC-OmniGS: Self-Calibrating Omnidirectional Gaussian Splatting (ICLR 2025 · similarity 0.86)
- UniK3D: Universal Camera Monocular 3D Estimation (CVPR 2025 · similarity 0.86)
- 360Loc: A Dataset and Benchmark for Omnidirectional Visual Localization with Cross-device Queries (CVPR 2024 · similarity 0.85)
- Multi-view Reconstruction via SfM-guided Monocular Depth Estimation (CVPR 2025 · similarity 0.85)
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.
