Research map / Detection and segmentation / CNN features
Learning to Detect Features in Texture Images
Linguang Zhang, Szymon Rusinkiewicz
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 peakedness-loss tuning to the ranking-trained detector (ablation) | Texture-specific feature detector · Carpet | repeatable features: 3290 -> 3715 | Improved |
| Add peakedness-loss tuning to the ranking-trained detector (ablation) | Texture-specific feature detector · Asphalt | repeatable features: 3332 -> 3344 | No clear effect |
| Add peakedness-loss tuning to the ranking-trained detector (ablation) | Texture-specific feature detector · Wood | repeatable features: 1892 -> 2906 | Improved |
| Add peakedness-loss tuning to the ranking-trained detector (ablation) | Texture-specific feature detector · Tile | repeatable features: 3383 -> 3397 | Mixed |
| Add peakedness-loss tuning to the ranking-trained detector (ablation) | Texture-specific feature detector · Granite | repeatable features: 3384 -> 3431 | No clear effect |
| Add peakedness-loss tuning to the ranking-trained detector (ablation) | Texture-specific feature detector · Concrete | repeatable features: 2716 -> 3607 | Improved |
| Add peakedness-loss tuning to the ranking-trained detector (ablation) | Texture-specific feature detector · Coarse | repeatable features: 3188 -> 3314 | Improved |
| Set peakedness-loss weight α above 0.5 (ablation) | Tuned detector · Texture images | repeatability gain | Hurt |
| Use peakedness window w=7 rather than other tested window sizes (ablation) | Tuned detector · Texture images | repeatability gain | Improved |
| Use k=10 or k=20 scores to compute peakedness (ablation) | Tuned detector · Texture images | repeatability gain | Improved |
| Train a universal detector on the union of texture images, using ranking loss only (ablation) | Feature detector · Cross-texture evaluation | repeatability | Hurt |
| Train the detector on granite rather than the union of textures (ablation) | Feature detector · Cross-texture evaluation | repeatability | Improved |
Closest papers by meaning
- Deep Filter Banks for Texture Recognition, Description, and Segmentation (IJCV 2016 · similarity 0.88)
- Perceptual Texture Similarity Estimation: An Evaluation of Computational Features (TPAMI 2020 · similarity 0.85)
- LoFTR: Detector-Free Local Feature Matching With Transformers (CVPR 2021 · similarity 0.85)
- Visualizing and Understanding Deep Texture Representations (CVPR 2016 · similarity 0.85)
- ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression (NeurIPS 2025 · similarity 0.84)
- Recognize Any Regions (NeurIPS 2024 · 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.
