Research map / Detection and segmentation / Open-set and few-shot detection
Snap Angle Prediction for 360° Panoramas
Bo Xiong, Kristen Grauman
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 pixel-objectness foreground objective with learned snap angle prediction (vs Canonical view) | snap angle model · 360 panoramas, all | object overlap IoU: 74.4% -> 81.1% | Improved |
| add learned snap angles instead of random rotation | snap angle model · User study, all | preference rate: Prefer Ours 65.3% vs Random 22.4% | Improved |
| add learned snap angles instead of canonical orientation | snap angle model · User study, all | preference rate: Prefer Ours 53.8% vs Canonical 31.5% | Improved |
| add snap angles to cubemaps for memorability | snap angle model · Image memorability (all, normalized) | normalized score: 46.8% -> 59.8% | Improved |
| add snap angles to cubemaps for aesthetics | snap angle model · Image aesthetics (all, normalized) | normalized score: 44.3% -> 52.9% | Improved |
| project panoramas with snap angles before feeding pretrained classifier (single face) | ResNet-101 · Cubemap recognition, Single | accuracy (%): 68.5 -> 70.1 | Improved |
| project panoramas with snap angles before feeding pretrained classifier (pano) | ResNet-101 · Cubemap recognition, Pano | accuracy (%): 66.5 -> 68.1 | Improved |
| replace pixel objectness input with ground-truth foreground maps (T=4) (ablation) | snap angle model · Foreground disruption | foreground disruption error: 0.283 -> 0.259 | Improved |
| increase search budget T from 2 to 10 with pixel objectness input (ablation) | snap angle model · Foreground disruption | foreground disruption error: 0.305 -> 0.277 | Improved |
| increase search budget T from 2 to 10 with ground-truth input (ablation) | snap angle model · Foreground disruption | foreground disruption error: 0.274 -> 0.234 | Improved |
| use learned content-based search instead of Uniform rotate | snap angle model · Foreground disruption vs budget | foreground disruption: up to 10% (absolute) | Improved |
| use learned content-based search instead of Saliency baseline | snap angle model · Foreground disruption vs budget | foreground disruption: up to 31% (absolute) | Improved |
Closest papers by meaning
- CubeDiff: Repurposing Diffusion-Based Image Models for Panorama Generation (ICLR 2025 · similarity 0.86)
- HORIZON: High-Resolution Semantically Controlled Panorama Synthesis (AAAI 2024 · similarity 0.86)
- Splatter-360: Generalizable 360 Gaussian Splatting for Wide-baseline Panoramic Images (CVPR 2025 · similarity 0.84)
- PanSplat: 4K Panorama Synthesis with Feed-Forward Gaussian Splatting (CVPR 2025 · similarity 0.84)
- LEAP: Liberate Sparse-View 3D Modeling from Camera Poses (ICLR 2024 · similarity 0.84)
- TAPVid-360: Tracking Any Point in 360 from Narrow Field of View Video (NeurIPS 2025 · similarity 0.84)
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.
