# OmniDS: Dual-Stream Context Fusion for Omnidirectional Depth from Fisheye Cameras

arXiv: https://arxiv.org/abs/2607.03038 · HTML: https://mlsanity.com/paper/2607.03038/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 24 of its 33 references that are in the map.
- First posted on arXiv 3 Jul 2026.

## Builds on (most cited first, 24 of 24)

- [Emerging Properties in Self-Supervised Vision Transformers](https://mlsanity.com/paper/2104.14294/) (2021)
- [Deformable DETR: Deformable Transformers for End-to-End Object Detection](https://mlsanity.com/paper/2010.04159/) (2020)
- [RAFT: Recurrent All-Pairs Field Transforms for Optical Flow](https://mlsanity.com/paper/2003.12039/) (2020)
- [DINOv3](https://mlsanity.com/paper/2508.10104/) (2025)
- [LoFTR: Detector-Free Local Feature Matching with Transformers](https://mlsanity.com/paper/2104.00680/) (2021)
- [KITTI-360: A Novel Dataset and Benchmarks for Urban Scene Understanding in 2D and 3D](https://mlsanity.com/paper/2109.13410/) (2021)
- [Intriguing Properties of Vision Transformers](https://mlsanity.com/paper/2105.10497/) (2021)
- [RAFT-Stereo: Multilevel Recurrent Field Transforms for Stereo Matching](https://mlsanity.com/paper/2109.07547/) (2021)
- [How Do Vision Transformers Work?](https://mlsanity.com/paper/2202.06709/) (2022)
- [RoMa: Robust Dense Feature Matching](https://mlsanity.com/paper/2305.15404/) (2023)
- [A Tale of Two Features: Stable Diffusion Complements DINO for Zero-Shot Semantic Correspondence](https://mlsanity.com/paper/2305.15347/) (2023)
- [Group-wise Correlation Stereo Network](https://mlsanity.com/paper/1903.04025/) (2019)
- [Iterative Geometry Encoding Volume for Stereo Matching](https://mlsanity.com/paper/2303.06615/) (2023)
- [FoundationStereo: Zero-Shot Stereo Matching](https://mlsanity.com/paper/2501.09898/) (2025)
- [WoodScape: A multi-task, multi-camera fisheye dataset for autonomous driving](https://mlsanity.com/paper/1905.01489/) (2019)
- [Attention Concatenation Volume for Accurate and Efficient Stereo Matching](https://mlsanity.com/paper/2203.02146/) (2022)
- [MatchFormer: Interleaving Attention in Transformers for Feature Matching](https://mlsanity.com/paper/2203.09645/) (2022)
- [Improving Vision Transformers by Revisiting High-frequency Components](https://mlsanity.com/paper/2204.00993/) (2022)
- [Eliminating the Blind Spot: Adapting 3D Object Detection and Monocular Depth Estimation to 360deg Panoramic Imagery](https://mlsanity.com/paper/1808.06253/) (2018)
- [SweepNet: Wide-baseline Omnidirectional Depth Estimation](https://mlsanity.com/paper/1902.10904/) (2019)
- [OmniMVS: End-to-End Learning for Omnidirectional Stereo Matching](https://mlsanity.com/paper/1908.06257/) (2019)
- [RomniStereo: Recurrent Omnidirectional Stereo Matching](https://mlsanity.com/paper/2401.04345/) (2024)
- [EDM: Equirectangular Projection-Oriented Dense Kernelized Feature Matching](https://mlsanity.com/paper/2502.20685/) (2025)
- [Helvipad: A Real-World Dataset for Omnidirectional Stereo Depth Estimation](https://mlsanity.com/paper/2411.18335/) (2024)

## Abstract (arXiv)

Omnidirectional depth estimation from multi-fisheye camera rigs is complicated by visibility conflicts: wide baselines cause different cameras to observe different portions, or even different faces, of the same object, so aggregating their features into a unified equirectangular (ERP) representation under fixed projection produces ambiguous matching evidence near occlusion boundaries and thin structures. Although existing methods mitigate this by down-weighting unreliable views, they do not resolve the underlying discrepancy because context formation and cross-view fusion remain tied to rigid fisheye-to-ERP sampling. We present OmniDS, an iterative depth refinement framework that replaces rigid aggregation by combining dynamic context fusion with consensus-aware multi-view similarity. A dual-stream encoder pairs a lightweight CNN for geometric detail with a frozen DINOv3 for semantic priors; their features are reprojected into ERP space at each refinement step via learned view weighting and deformable cross-attention with geometric distortion bias. In parallel, a multi-view consensus volume captures global cross-camera agreement through group-wise correlation and feature variance, regularized by a 3D U-Net. For efficient deployment, we distill the dual-stream representation into a single MobileNet-based encoder. OmniDS achieves state-of-the-art performance on the OmniThings, OmniHouse, and Sunny benchmarks while maintaining competitive inference speed. Project page and codes are available at https://parkchaesong.github.io/omnids.