Research map / Detection and segmentation / CNN features
LRANet++: Low-Rank Approximation Network for Accurate and Efficient Text Spotting
Yuchen Su, Zhineng Chen, Yongkun Du, Zuxuan Wu, Hongtao Xie, Yu–Gang Jiang
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 |
|---|---|---|---|
| use contour-driven instead of mask-driven representation (ablation) | LRA · CTW1500 | E2E F-measure: 65.1 -> 67.7 | Improved |
| use contour-driven instead of side-driven representation (ablation) | LRA · CTW1500 | E2E F-measure: 67.3 -> 67.7 | Improved |
| increase orthanchor dimension from 10 to 14 (ablation) | LRANet++ · CTW1500 | IoU: 95.8 -> 98.1 | Improved |
| increase orthanchor dimension from 14 to 18 (ablation) | LRANet++ · CTW1500 | detection F-measure: 87.2 -> 87.1 | No clear effect |
| use FMS instead of SVD under 20% label noise (ablation) | LRA projection · CTW1500 | IoU: 93.6 -> 97.2 | Improved |
| use FMS instead of SVD at dimension 6 (ablation) | LRA projection · CTW1500 curved subset | IoU: 85.3 -> 87.8 | Improved |
| use FMS instead of SVD at dimension 22 (ablation) | LRA projection · CTW1500 curved subset | detection F-measure: 84.3 -> 84.4 | No clear effect |
| use triple assignment instead of single sparse (ablation) | detection head · Total-Text | E2E F-measure: 81.1 -> 82.2 | Improved |
| use triple assignment instead of dual assignment (ablation) | detection head · Total-Text | inference time (ms): 6.5 -> 3.2 | Improved |
| use CTC decoding instead of autoregressive (ablation) | recognition head · CTW1500 | inference time (ms): 33.2 -> 15.1 | Mixed |
| use proposed recognition head instead of ABCNet v2 head (ablation) | LRANet++ · CTW1500 | E2E F-measure: 65.1 -> 67.7 | Improved |
| use large-ratio image scaling augmentation [0.1, 3] (ablation) | LRANet++ · Total-Text | E2E F-measure: 80.3 -> 82.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SwinTextSpotter v2 — SwinTextSpotter v2: Towards Better Synergy for Scene Text Spotting (IJCV 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LRANet: Towards Accurate and Efficient Scene Text Detection with Low-Rank Approximation Network (AAAI 2024 · similarity 0.94)
- PAN++: Towards Efficient and Accurate End-to-End Spotting of Arbitrarily-Shaped Text (TPAMI 2021 · similarity 0.90)
- TextDragon: An End-to-End Framework for Arbitrary Shaped Text Spotting (ICCV 2019 · similarity 0.90)
- ABCNet v2: Adaptive Bezier-Curve Network for Real-time End-to-end Text Spotting (TPAMI 2021 · similarity 0.90)
- Bridging the Gap Between End-to-End and Two-Step Text Spotting (CVPR 2024 · similarity 0.89)
- SemiETS: Integrating Spatial and Content Consistencies for Semi-Supervised End-to-end Text Spotting (CVPR 2025 · similarity 0.88)
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.
