Research map / Detection and segmentation / CNN features
HexNet: An Orientation-Aware Deep Learning Framework for Omni-Directional Input
Chao Zhang, Stephan Liwicki, Sen He, William A. P. Smith, Roberto Cipolla
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 arc-based north-aligned interpolation (ablation) | Hex-UNet · Synthia-O (r=7) | mIoU: 45.1 -> 49.5 | Improved |
| initialise from perspective U-Net filters with 10 epochs refinement (Hex-UNet-T) instead of training from scratch | Hex-UNet · Synthia-O (r=6) | mIoU: 43.6 -> 36.7 | Hurt |
| filter transfer at higher resolution | Hex-UNet-T · Synthia-O (r=8) | mIoU: 45.3 vs 47.1 | Mixed |
| use spherical instead of planar representation | U-Net · Synthia-S (lv-8 vs planar 1/3) | mIoU: 56.5 -> 57.6 | Improved |
| use spherical instead of planar representation | U-Net · Cityscapes (1/6 res) | mIoU: 51.5 -> 54.3 | Improved |
| use spherical representation with ImageNet pretraining | DANet · Cityscapes (1/6 res) | mIoU: 52.5 -> 56.6 | Improved |
| add ImageNet pretraining to planar DANet | DANet · Cityscapes (1/6 res) | mIoU: 51.2 -> 57.9 | Improved |
| replace 3x3 kernel with hexagonal kernel (no sphere) (ablation) | U-Net · Cityscapes (1/3 res) | mIoU: 55.5 -> 55.2 | No clear effect |
| replace 3x3 kernel with hexagonal kernel (no sphere), pretrained (ablation) | DANet · Cityscapes (1/6 res) | mIoU: 57.9 -> 57.1 | Hurt |
| raise mesh resolution from r=6 to r=7 (ablation) | Hex-UNet · Synthia-O | mIoU: 43.6 -> 49.5 | Improved |
| raise mesh resolution from r=7 to r=8 (ablation) | Hex-UNet · Synthia-O | mIoU: 49.5 -> 47.1 | Hurt |
| use masked partial-sphere computation (ablation) | Graph-HexNet · Synthia-O (r=7) | memory (MB): 3,596 -> 595 | 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.
- spherePHD — SpherePHD: Applying CNNs on a Spherical PolyHeDron Representation of 360deg Images (CVPR 2019) · 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
- SphereNet: Learning Spherical Representations for Detection and Classification in Omnidirectional Images (ECCV 2018 · similarity 0.87)
- Learning to Orient Surfaces by Self-supervised Spherical CNNs (NeurIPS 2020 · similarity 0.87)
- DeepSphere: a graph-based spherical CNN (ICLR 2020 · similarity 0.86)
- Unified Spherical Frontend: Learning Rotation-Equivariant Representations of Spherical Images from Any Camera (CVPR 2026 · similarity 0.86)
- SpherePHD: Applying CNNs on a Spherical PolyHeDron Representation of 360deg Images (CVPR 2019 · similarity 0.86)
- Rotation Equivariant Graph Convolutional Network for Spherical Image Classification (CVPR 2020 · 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.
