Research map / Detection and segmentation / CNN features
Few Could Be Better Than All: Feature Sampling and Grouping for Scene Text Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use feature sampling at all three scales (64+128+256 features) instead of sampling only at L0 or L0+L1 (ablation) | Feature sampling and grouping detector · ICDAR 2015 test | F-measure: 67.0 -> 89.1; 82.2 -> 89.1 | Helped |
| Use feature sampling at all three scales (64+128+256 features) instead of sampling only at L0 or L0+L1 (ablation) | Feature sampling and grouping detector · MLT-2017 val | F-measure: 63.9 -> 79.5; 73.4 -> 79.5 | Helped |
| Increase sampled features from 16+32+64 to 32+64+128 (ablation) | Feature sampling and grouping detector · ICDAR 2015 test | F-measure: 77.8 -> 86.0 | Helped |
| Increase sampled features from 32+64+128 to 64+128+256 (ablation) | Feature sampling and grouping detector · ICDAR 2015 test | F-measure: 86.0 -> 89.1 | Helped |
| Increase sampled features from 64+128+256 to 128+256+512 (ablation) | Feature sampling and grouping detector · ICDAR 2015 test | F-measure: 89.1 -> 89.0 | Hurt |
| Use adaptive sampling of 25% of foreground features (ablation) | Feature sampling and grouping detector · ICDAR 2015 test | F-measure: 89.1 -> 88.9 | No clear effect |
| Replace transformer grouping with fully connected layers (ablation) | Feature sampling and grouping detector · ICDAR 2015 test | F-measure: 89.1 -> 85.7 | Hurt |
| Replace transformer grouping with GCN (ablation) | Feature sampling and grouping detector · MTWI test | F-measure: 75.2 -> 72.5 | Hurt |
| Add constrained deformable pooling instead of average pooling (ablation) | Feature sampling and grouping detector · ICDAR 2015 test | F-measure: 88.3 -> 89.1 | Helped |
| Add constrained deformable pooling instead of standard deformable pooling (ablation) | Feature sampling and grouping detector · MLT-2017 val | F-measure: 79.2 -> 79.5 | Helped |
| Use scale-invariant GWD loss instead of original GWD loss (ablation) | Feature sampling and grouping detector · ICDAR 2015 test | F-measure: 88.4 -> 89.1 | Helped |
| Use scale-invariant GWD loss instead of original GWD loss (ablation) | Feature sampling and grouping detector · MLT-2017 val | F-measure: 79.0 -> 79.5 | Helped |
Closest papers by meaning
- Scene Text Detection with Supervised Pyramid Context Network (AAAI 2019 · similarity 0.89)
- EAST: An Efficient and Accurate Scene Text Detector (CVPR 2017 · similarity 0.88)
- Kernel Adaptive Convolution for Scene Text Detection via Distance Map Prediction (CVPR 2024 · similarity 0.87)
- Text Grouping Adapter: Adapting Pre-trained Text Detector for Layout Analysis (CVPR 2024 · similarity 0.86)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.85)
- LayoutFormer: Hierarchical Text Detection Towards Scene Text Understanding (CVPR 2024 · 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.