Research map / Detection and segmentation / Open-set and few-shot detection
Learned Shape-Tailored Descriptors for Segmentation
Naeemullah Khan, Ganesh Sundaramoorthi
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 |
|---|---|---|---|
| learn an invariant metric over shape-tailored descriptors | texture segmentation · Real-World Texture Dataset | GT-cov. ODS: 0.86 -> 0.92 | Improved |
| use shape-tailored descriptors rather than non-STLD descriptors (ablation) | learned descriptor segmentation · Real-World Texture Dataset | GT-cov. ODS: 0.89 -> 0.92 | Improved |
| add learned invariance to handcrafted STLD (ablation) | texture segmentation · Real-World Texture Dataset | GT-cov. ODS: 0.86 -> 0.92 | Improved |
| use a 5by5, 4by4, or 3by3 box tessellation initialization (ablation) | Learned STLD · Real-World Texture Dataset | GT-cov. ODS: 0.91; 0.91; 0.91 | No clear effect |
| vary hidden units from 21 to 61 in the two-layer network (ablation) | Learned STLD · Real-World Texture Dataset | GT-cov. ODS: 0.90; 0.91; 0.91; 0.91; 0.91 | No clear effect |
| increase network depth to 3 or 4 layers with 41 units (ablation) | Learned STLD · Real-World Texture Dataset | GT-cov. ODS: 0.91 -> 0.89; 0.91 -> 0.88 | Hurt |
| reduce training images from 128 to 100, 75, or 50 (ablation) | Learned STLD · Real-World Texture Dataset | GT-cov. ODS: 0.91 -> 0.90; 0.91 -> 0.90; 0.91 -> 0.89 | Hurt |
| vary the number of dilation steps during training (ablation) | Learned STLD · Real-World Texture Dataset | GT-cov. ODS: 0.91; 0.91; 0.90; 0.91; 0.90 | No clear effect |
| use learned shape-tailored descriptors instead of handcrafted STLD | texture segmentation · Synthetic Texture Dataset | GT-cov. ODS: 0.87 -> 0.90 | Improved |
| use learned shape-tailored descriptors instead of handcrafted STLD | texture segmentation · Graz | GT-cov. ODS: 0.70 -> 0.76 | Improved |
| use learned shape-tailored descriptors instead of handcrafted STLD | texture segmentation · BSDS500 | GT-cov. ODS: 0.57 -> 0.67 | Improved |
| use learned shape-tailored descriptors rather than VGG convolutional descriptors | texture segmentation · Real-World Texture Dataset | GT-cov. ODS: 0.84 -> 0.92; 0.79 -> 0.92; 0.86 -> 0.92 | Improved |
Closest papers by meaning
- Deep Filter Banks for Texture Recognition, Description, and Segmentation (IJCV 2016 · similarity 0.87)
- Visualizing and Understanding Deep Texture Representations (CVPR 2016 · similarity 0.87)
- Shape or Texture: Understanding Discriminative Features in CNNs (ICLR 2021 · similarity 0.86)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.85)
- ShapeEmbed: a self-supervised learning framework for 2D contour quantification (NeurIPS 2025 · similarity 0.85)
- Visual Anagrams Reveal Hidden Differences in Holistic Shape Processing Across Vision Models (NeurIPS 2025 · 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.
