Research map / Detection and segmentation / CNN features
SphereNet: Learning Spherical Representations for Detection and Classification in Omnidirectional Images
Benjamin Coors, Alexandru Paul Condurache, Andreas Geiger
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 |
|---|---|---|---|
| replace regular convolution with SphereNet kernels (bilinear interpolation) | EquirectCNN · Omni-MNIST | test error (%): 9.61 -> 5.59 | Improved |
| use nearest neighbor instead of bilinear interpolation in SphereNet (ablation) | SphereNet · Omni-MNIST | test error (%): 5.59 -> 7.03 | Hurt |
| use uniform sphere sampling instead of equirectangular grid (NN) (ablation) | SphereNet · Omni-MNIST | test error (%): 7.03 -> 7.16 | No clear effect |
| use uniform sphere sampling instead of equirectangular grid (NN) (ablation) | SphereNet · Omni-MNIST | number of parameters: 196K -> 144K | Improved |
| add Spherical Transformer Network to EquirectCNN | EquirectCNN · Omni-MNIST | test error (%): 9.61 -> 8.22 | Improved |
| add Spherical Transformer Network to EquirectCNN | EquirectCNN · FlyingCars test | mAP (%): 41.57 -> 38.91 | Hurt |
| use cube map representation instead of equirectangular input | CNN · Omni-MNIST | test error (%): 9.61 -> 10.03 | Hurt |
| use cube map representation instead of equirectangular input | CNN · FlyingCars test | mAP (%): 41.57 -> 48.42 | Improved |
| enlarge kernel size towards the poles (EquirectCNN++) | EquirectCNN · FlyingCars test | mAP (%): 41.57 -> 45.65 | Improved |
| use SphereNet (NN) instead of cube map CNN | Sphere-SSD · FlyingCars test | mAP (%): 48.42 -> 50.18 | Improved |
| use SphereNet (NN) instead of cube map CNN when transferring a KITTI-trained detector | Sphere-SSD · OmPaCa test (transfer from KITTI) | mAP (%): 34.19 -> 49.73 | Improved |
| use SphereNet (NN) instead of equirectangular CNN when transferring a KITTI-trained detector | Sphere-SSD · OmPaCa test (transfer from KITTI) | mAP (%): 43.43 -> 49.73 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GCNN — G-CNN: An Iterative Grid Based Object Detector (CVPR 2016) · better on 1 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
- SpherePHD: Applying CNNs on a Spherical PolyHeDron Representation of 360deg Images (CVPR 2019 · similarity 0.91)
- Geometry Aware Convolutional Filters for Omnidirectional Images Representation (ICML 2019 · similarity 0.90)
- Learning Spherical Convolution for Fast Features from 360° Imagery (NeurIPS 2017 · similarity 0.90)
- SpherePHD: Applying CNNs on 360 Images With Non-Euclidean Spherical PolyHeDron Representation (TPAMI 2020 · similarity 0.90)
- Spherical DNNs and Their Applications in 360 Images and Videos (TPAMI 2021 · similarity 0.89)
- SphereUFormer: A U-Shaped Transformer for Spherical 360 Perception (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.
