# Learning Spherical Convolution for Fast Features from 360deg Imagery

arXiv: https://arxiv.org/abs/1708.00919 · HTML: https://mlsanity.com/paper/1708.00919/

## Key facts

- Cited by 22 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 2 Aug 2017.

## Built on by (most cited first, 22 of 22)

- [PanoFormer: Panorama Transformer for Indoor 360 Depth Estimation](https://mlsanity.com/paper/2203.09283/) (2022)
- [Geometric Deep Learning and Equivariant Neural Networks](https://mlsanity.com/paper/2105.13926/) (2021)
- [OmniFusion: 360 Monocular Depth Estimation via Geometry-Aware Fusion](https://mlsanity.com/paper/2203.00838/) (2022)
- [Deep Learning for Omnidirectional Vision: A Survey and New Perspectives](https://mlsanity.com/paper/2205.10468/) (2022)
- [Look at the Neighbor: Distortion-aware Unsupervised Domain Adaptation for Panoramic Semantic Segmentation](https://mlsanity.com/paper/2308.05493/) (2023)
- [BiFuse++: Self-supervised and Efficient Bi-projection Fusion for 360 Depth Estimation](https://mlsanity.com/paper/2209.02952/) (2022)
- [PanoSwin: a Pano-style Swin Transformer for Panorama Understanding](https://mlsanity.com/paper/2308.14726/) (2023)
- [Equivariant Networks for Pixelized Spheres](https://mlsanity.com/paper/2106.06662/) (2021)
- [GLPanoDepth: Global-to-Local Panoramic Depth Estimation](https://mlsanity.com/paper/2202.02796/) (2022)
- [Complementary Bi-directional Feature Compression for Indoor 360deg Semantic Segmentation with Self-distillation](https://mlsanity.com/paper/2207.02437/) (2022)
- [PanoDepth: A Two-Stage Approach for Monocular Omnidirectional Depth Estimation](https://mlsanity.com/paper/2202.01323/) (2022)
- [Single-Shot Cuboids: Geodesics-based End-to-end Manhattan Aligned Layout Estimation from Spherical Panoramas](https://mlsanity.com/paper/2102.03939/) (2021)
- [OSLO: On-the-Sphere Learning for Omnidirectional images and its application to 360-degree image compression](https://mlsanity.com/paper/2107.09179/) (2021)
- [Distortion-Tolerant Monocular Depth Estimation On Omnidirectional Images Using Dual-cubemap](https://mlsanity.com/paper/2203.09733/) (2022)
- [ASOD60K: An Audio-Induced Salient Object Detection Dataset for Panoramic Videos](https://mlsanity.com/paper/2107.11629/) (2021)
- [EAGOR: Embodied Reasoning in Omni-direction](https://mlsanity.com/paper/2607.06165/) (2026)
- [Interpolated SelectionConv for Spherical Images and Surfaces](https://mlsanity.com/paper/2210.10123/) (2022)
- [SphereFusion: Efficient Panorama Depth Estimation via Gated Fusion](https://mlsanity.com/paper/2502.05859/) (2025)
- [Spherical Convolution empowered FoV Prediction in 360-degree Video Multicast with Limited FoV Feedback](https://mlsanity.com/paper/2201.12525/) (2022)
- [Extracting the gamma-ray source-count distribution below the Fermi-LAT detection limit with deep learning](https://mlsanity.com/paper/2302.01947/) (2023)
- [WinDB: HMD-free and Distortion-free Panoptic Video Fixation Learning](https://mlsanity.com/paper/2305.13901/) (2023)
- [Distortion-Aware Loop Filtering of Intra 360^o Video Coding with Equirectangular Projection](https://mlsanity.com/paper/2202.09802/) (2022)

## Abstract (arXiv)

While 360deg cameras offer tremendous new possibilities in vision, graphics, and augmented reality, the spherical images they produce make core feature extraction non-trivial. Convolutional neural networks (CNNs) trained on images from perspective cameras yield "flat" filters, yet 360deg images cannot be projected to a single plane without significant distortion. A naive solution that repeatedly projects the viewing sphere to all tangent planes is accurate, but much too computationally intensive for real problems. We propose to learn a spherical convolutional network that translates a planar CNN to process 360deg imagery directly in its equirectangular projection. Our approach learns to reproduce the flat filter outputs on 360deg data, sensitive to the varying distortion effects across the viewing sphere. The key benefits are 1) efficient feature extraction for 360deg images and video, and 2) the ability to leverage powerful pre-trained networks researchers have carefully honed (together with massive labeled image training sets) for perspective images. We validate our approach compared to several alternative methods in terms of both raw CNN output accuracy as well as applying a state-of-the-art "flat" object detector to 360deg data. Our method yields the most accurate results while saving orders of magnitude in computation versus the existing exact reprojection solution.