Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Gabor Texture Features for Fine-Grained Recognition
Lanyun Zhu, Tianrun Chen, Jianxiong Yin, Simon See, Jun Liu
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 Gabor texture branch and region selection gate (ablation) | ResNet50 · CUB-200-2011 | accuracy (%): 84.5 -> 90.8 | Improved |
| add region selection gate to texture branch (ablation) | texture branch · CUB-200-2011 | accuracy (%) and FLOPs: 89.0 -> 90.8; 36.88G -> 20.72G FLOPs | Improved |
| use Gabor filters instead of 3x3 convolutions (ablation) | texture branch · CUB-200-2011 | accuracy (%): 85.9 -> 90.8 | Improved |
| learn Gabor parameters in supervised manner instead of manual setting (ablation) | Gabor filters · CUB-200-2011 | accuracy (%): 86.5 -> 90.8 | Improved |
| add parameter value constraints to learnable Gabor filters (ablation) | Gabor filters · CUB-200-2011 | accuracy (%): 75.4 -> 90.8 | Improved |
| add high frequency enhancement strategy (ablation) | Gabor filters · CUB-200-2011 | accuracy (%): 88.1 -> 90.8 | Improved |
| add Learnable Histogram Operator (ablation) | statistical feature extractor · CUB-200-2011 | accuracy (%): 87.0 -> 90.8 | Improved |
| add Filter Correlation Module (ablation) | statistical feature extractor · CUB-200-2011 | accuracy (%): 89.2 -> 90.8 | Improved |
| use LHO instead of hand-crafted statistics (ablation) | texture branch · CUB-200-2011 | accuracy (%): 87.9 -> 90.8 | Improved |
| place LHO on Gabor texture branch rather than CNN branch (ablation) | LHO · CUB-200-2011 | accuracy (%): 86.2 -> 90.8 | Improved |
| apply method to ImageNet instead of fine-grained data | ResNet50 · ImageNet | top-1 accuracy (%): 76.1 -> 77.9 | Improved |
| vary loss weight lambda from 0.01 to 1 (ablation) | full method · CUB-200-2011 | accuracy (%): 90.0 to 90.8 | No clear effect |
Closest papers by meaning
- Learning a Discriminative Filter Bank Within a CNN for Fine-Grained Recognition (CVPR 2018 · similarity 0.88)
- A Lightweight Hybrid Gabor Deep Learning Approach and its Application to Medical Image Classification (IJCV 2026 · similarity 0.87)
- Deep Filter Banks for Texture Recognition, Description, and Segmentation (IJCV 2016 · similarity 0.87)
- Fine-Grained Recognition: Accounting for Subtle Differences between Similar Classes (AAAI 2020 · similarity 0.87)
- From BoW to CNN: Two Decades of Texture Representation for Texture Classification (IJCV 2018 · similarity 0.86)
- From Few-way to Many-way: Rethinking Few-shot Fine-grained Image Classification (CVPR 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.
