Research map / 3D vision / Face and shape
Scene Categorization From Contours: Medial Axis Based Salience Measures
Morteza Rezanejad, Gabriel Downs, John Wilder, Dirk B. Walther, Allan Jepson, Sven Dickinson, Kaleem Siddiqi
Check 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 medial-axis salience-weighted contour channels to intact contours | Pre-trained CNN features with linear SVM · Artist Scenes | top-1 accuracy: Contours: 86.53; Contours, Ribbon, Separ.: 94.96 | Improved |
| Add ribbon-salience weighted contour channel | Pre-trained CNN features with linear SVM · MIT67 | top-1 accuracy: 26.36 -> 27.18 (VGG16); 23.88 -> 25.74 (Res50) | Improved |
| Add taper-salience weighted contour channel | Pre-trained CNN features with linear SVM · Places365 | top-1 accuracy: 8.02 -> 11.73 | Improved |
| Add separation-salience weighted contour channel | Pre-trained CNN features with linear SVM · Places365 | top-1 accuracy: 8.02 -> 10.53 | Improved |
| Combine ribbon and separation salience channels with intact contours | Pre-trained CNN features with linear SVM · Places365 | top-1 accuracy: 8.02 -> 14.23 | Improved |
| Replace the intact contour channel with all three salience-weighted channels (ablation) | Pre-trained CNN features with linear SVM · MIT67 | top-1 accuracy: Contours 26.36; Ribbon, Taper, Separ. 25.84 (VGG16) | Hurt |
| Use ribbon plus taper instead of ribbon plus separation salience (ablation) | Contours plus salience channels, VGG16 · Artist Scenes | top-1 accuracy: Contours, Ribbon, Taper 94.53; Contours, Ribbon, Separ. 94.96 | Hurt |
| Use ribbon plus taper instead of taper plus separation salience (ablation) | Contours plus salience channels, ResNet50 · MIT67 | top-1 accuracy: Contours, Ribbon, Taper 25.44; Contours, Taper, Separ. 30.37 | Hurt |
| Use all three salience channels without intact contours (ablation) | Pre-trained CNN features with linear SVM · Places365 | top-1 accuracy: Contours, Ribbon, Separ 14.23; Ribbon, Taper, Separ 12.64 | Hurt |
| Select the top 50% rather than bottom 50% of contour pixels by salience (ablation) | CNN scene classification · Artist Scenes and MIT67 | top-1 accuracy | Improved |
| Select the top 50% rather than bottom 50% of contour pixels by salience (ablation) | Human scene categorization · Artist Scenes | accuracy | Improved |
Closest papers by meaning
- Shape-Based Measures Improve Scene Categorization (TPAMI 2023 · similarity 0.93)
- Sketch2Saliency: Learning To Detect Salient Objects From Human Drawings (CVPR 2023 · similarity 0.88)
- Shape or Texture: Understanding Discriminative Features in CNNs (ICLR 2021 · similarity 0.87)
- ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression (NeurIPS 2025 · similarity 0.86)
- Visual Anagrams Reveal Hidden Differences in Holistic Shape Processing Across Vision Models (NeurIPS 2025 · similarity 0.86)
- Vision Transformers Learn Gestalt-Like Figure-Ground Cues from Natural Images (NeurIPS 2026 · similarity 0.85)
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.
