Research map / 3D vision / Depth and stereo
Beyond Image Borders: Learning Feature Extrapolation for Unbounded Image Composition
Xiaoyu Liu, Ming Liu, Junyi Li, Shuai Liu, Xiaotao Wang, Lei Lei, Wangmeng Zuo
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 feature-space extrapolation module (FEM) to the unbounded regression model (ablation) | UNIC · GAICD unbounded, Acc1/5 eps=0.85 | Acc_1/5 (%): 48.1 -> 59.0 | Improved |
| add feature-space extrapolation module (FEM) (ablation) | UNIC · GAICD unbounded, Acc1/5 eps=0.90 | Acc_1/5 (%): 22.6 -> 23.2 | Improved |
| use image-level out-painting (SRN) instead of feature extrapolation (ablation) | UNIC · GAICD unbounded, Acc1/5 eps=0.90 | Acc_1/5 (%): 22.6 -> 19.9 | Hurt |
| replace feature extrapolation with image extrapolation by stronger generative models (QueryOTR) (ablation) | UNIC · GAICD unbounded, Acc1/5 eps=0.85 | Acc_1/5 (%): 48.1 -> 53.7 | Improved |
| use smooth-L1 rather than MSE for the extrapolation loss (ablation) | FEM · GAICD unbounded, Acc1/5 eps=0.85 | Acc_1/5 (%): 56.1 -> 59.0 | Improved |
| use KL-divergence instead of smooth-L1 for extrapolation loss (ablation) | FEM · GAICD unbounded, Acc1/5 eps=0.90 | Acc_1/5 (%): 23.2 -> 23.8 (eps=0.90), 59.0 -> 56.4 (eps=0.85) | Mixed |
| increase number of view adjustment steps from 1 to 3 (ablation) | UNIC · GAICD (83 large images), Acc1/5 eps=0.85 | Acc_1/5 (%): 48.2 -> 54.2 | Improved |
| increase adjustment steps from 2 to 3 (ablation) | UNIC · GAICD (83 large images), Acc1/5 eps=0.90 | Acc_1/5 (%): 19.3 -> 19.3 | No clear effect |
| apply UNIC (without FEM) to standard image cropping | UNIC · GAICD cropping, Acc1/5 eps=0.85 | Acc_1/5 (%): 85.0 -> 87.2 | Improved |
| UNIC vs CACNet on FLMS cropping | UNIC · FLMS cropping | IoU: 0.854 -> 0.840 | Hurt |
Closest papers by meaning
- Wide-Context Semantic Image Extrapolation (CVPR 2019 · similarity 0.87)
- TransView: Inside, Outside, and Across the Cropping View Boundaries (ICCV 2021 · similarity 0.86)
- Good View Hunting: Learning Photo Composition From Dense View Pairs (CVPR 2018 · similarity 0.85)
- Learning Subject-Aware Cropping by Outpainting Professional Photos (AAAI 2024 · similarity 0.83)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.83)
- Venus: Benchmarking and Empowering Multimodal Large Language Models for Aesthetic Guidance and Cropping (CVPR 2026 · similarity 0.83)
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.
