Research map / 3D vision / Depth and stereo
OmniFusion: 360 Monocular Depth Estimation via Geometry-Aware Fusion
Check 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 |
|---|---|---|---|
| Add geometry-aware feature fusion to the baseline (ablation) | OmniFusion · Stanford2D3D | Abs Rel: 9.7% | Helped |
| Add the transformer to the baseline (ablation) | OmniFusion · Stanford2D3D | Abs Rel: 5.7% | Helped |
| Add geometry-aware fusion and the transformer (ablation) | OmniFusion · Stanford2D3D, 1 iteration | Abs Rel: 15.4% | Helped |
| Add geometry-aware fusion and the transformer (ablation) | OmniFusion · Stanford2D3D, 2 iterations | Abs Rel: 16.4% | Helped |
| Increase refinement from 1 iteration to 2 (ablation) | OmniFusion · Stanford2D3D, Matterport3D, and 360D | Abs Rel | Helped |
| Increase refinement from 2 iterations to 3 (ablation) | OmniFusion · Stanford2D3D and Matterport3D | Abs Rel | Helped |
| Increase refinement from 3 iterations to 4 (ablation) | OmniFusion · Stanford2D3D and Matterport3D | Abs Rel | No clear effect |
| Increase the number of patches to 26 or more (ablation) | OmniFusion · Stanford2D3D | depth estimation performance | Hurt |
| Use patches that are too small, reducing their field of view (ablation) | OmniFusion · Stanford2D3D | depth estimation performance | Hurt |
| Use ResNet34 instead of ResNet18 as image encoder (ablation) | OmniFusion · Stanford2D3D | depth estimation performance | Helped |
| Use transformer depth 4 and 4 heads instead of depth 2 and 2 heads (ablation) | OmniFusion · Stanford2D3D, transformer ablation | Abs Rel: 0.1091 -> 0.1016 | Helped |
| Use transformer depth 8 and 8 heads instead of depth 6 and 4 heads (ablation) | OmniFusion · Stanford2D3D, transformer ablation | Abs Rel: 0.1026 -> 0.1044 | Hurt |
Closest papers by meaning
- OmniVGGT: Omni-Modality Driven Visual Geometry Grounded Transformer (CVPR 2026 · similarity 0.89)
- PatchFusion: An End-to-End Tile-Based Framework for High-Resolution Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.88)
- Single View Stereo Matching (CVPR 2018 · similarity 0.88)
- Depth Anywhere: Enhancing 360 Monocular Depth Estimation via Perspective Distillation and Unlabeled Data Augmentation (NeurIPS 2024 · similarity 0.87)
- RPG360: Robust 360 Depth Estimation with Perspective Foundation Models and Graph Optimization (NeurIPS 2025 · similarity 0.87)
- HELVIPAD: A Real-World Dataset for Omnidirectional Stereo Depth Estimation (CVPR 2025 · 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.