Research map / Image generation / Image restoration
Dissecting Image Crops
Basile Van Hoorick, Carl Vondrick
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 patch-based branch to the global thumbnail model (ablation) | Global crop detector · Flickr crop detection | accuracy: 79% -> 86% | Improved |
| add global thumbnail branch to the patch-based model (ablation) | Patch crop detector · Flickr crop detection | accuracy: 77% -> 86% | Improved |
| use joint model instead of human judgment | Joint model · Flickr crop detection | accuracy: 67% -> 86% | Improved |
| restrict test images to those without humans or faces | Joint model · Flickr crop detection | accuracy: 86% -> 81% | Hurt |
| add global branch to patch model on tilted photos (ablation) | Patch crop detector · Tilted test set | accuracy: 70% -> 71% | No clear effect |
| add global branch to patch model on texture photos (ablation) | Patch crop detector · Texture test set | accuracy: 67% -> 66% | No clear effect |
| add global branch to patch model on smooth white-wall photos (ablation) | Patch crop detector · Smooth test set | accuracy: 55% -> 50% | Hurt |
| keep color information (vs grayscale) for patch localization (ablation) | F_patch · Patch localization | accuracy: 15% -> 21% | Improved |
| keep color information (vs grayscale) for patch crop detector (ablation) | Patch crop detector · Flickr crop detection | accuracy: 72% -> 77% | Improved |
| keep color information (vs grayscale) for global crop detector (ablation) | Global crop detector · Flickr crop detection | accuracy: 78% -> 79% | No clear effect |
| replace cropping with rescaling to same dimensions (ablation) | Global crop detector · Flickr crop detection | accuracy: 79% -> 50% | Hurt |
| add artificial green transverse chromatic aberration to test images (ablation) | Patch localization / crop detection · Flickr test subset (3,500 photos) | accuracy | Improved |
Closest papers by meaning
- Image Cropping With Spatial-Aware Feature and Rank Consistency (CVPR 2023 · similarity 0.88)
- Cropper: Vision-Language Model for Image Cropping through In-Context Learning (CVPR 2025 · similarity 0.87)
- Venus: Benchmarking and Empowering Multimodal Large Language Models for Aesthetic Guidance and Cropping (CVPR 2026 · similarity 0.86)
- Learning Subject-Aware Cropping by Outpainting Professional Photos (AAAI 2024 · similarity 0.86)
- Spatial-Semantic Collaborative Cropping for User Generated Content (AAAI 2024 · similarity 0.85)
- Depth-Aware Concealed Crop Detection in Dense Agricultural Scenes (CVPR 2024 · 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.
