Research map / Vision-language and multimodal / Human and embodied modeling
Context Perception Parallel Decoder for Scene Text Recognition
Yongkun Du, Zhineng Chen, Caiyan Jia, Xiaoting Yin, Chenxia Li, Yuning Du, 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 |
|---|---|---|---|
| add character ordering module without side loss (ablation) | SVTR-B-CPPD · six English benchmarks | avg accuracy: 92.01 -> 92.97 | Improved |
| add character counting module without side loss (ablation) | SVTR-B-CPPD · six English benchmarks | avg accuracy: 92.01 -> 92.49 | Improved |
| add CC side loss to CC module (ablation) | SVTR-B-CPPD · six English benchmarks | avg accuracy: 92.49 -> 93.35 | Improved |
| add CO side loss to CO module (ablation) | SVTR-B-CPPD · six English benchmarks | avg accuracy: 92.97 -> 93.29 | Improved |
| add both modules and both side losses (ablation) | SVTR-B-CPPD · six English benchmarks | avg accuracy: 92.01 -> 93.80 | Improved |
| use CC loss instead of ACE loss (ablation) | SVTR-B-CPPD · six English benchmarks | avg accuracy: 93.31 -> 93.80 | Improved |
| use binary CO loss instead of standard CE (ablation) | SVTR-B-CPPD · six English benchmarks | avg accuracy: 93.50 -> 93.80 | Improved |
| add pre-decoding side loss to PD decoder (ablation) | SVTR-T-NRTR variants · six English benchmarks | avg accuracy: 89.60 -> 90.43 | Improved |
| replace AR decoder with standard PD decoder | SVTR-T-NRTR · six English benchmarks | avg accuracy: 91.49 -> 89.60 | Hurt |
| train AR decoder with permuted sequences (ablation) | SVTR-T-NRTR · six English benchmarks | avg accuracy: 91.49 -> 91.78 | Improved |
| remove linguistic self-attention from AR decoder (AR-L) (ablation) | SVTR-T-NRTR · six English benchmarks | avg accuracy: 91.49 -> 91.13 | No clear effect |
| replace CNN encoder NRTR with SVTR-T encoder | NRTR · six English benchmarks | avg accuracy: 89.62 -> 91.49 | Improved |
Closest papers by meaning
- One2Seq: One-Token Wise Decoder for Efficient Scene Text Recognition (AAAI 2026 · similarity 0.88)
- IPAD: Iterative, Parallel, and Diffusion-Based Network for Scene Text Recognition (IJCV 2025 · similarity 0.88)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.88)
- RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition (ECCV 2020 · similarity 0.88)
- Arbitrary Reading Order Scene Text Spotter with Local Semantics Guidance (AAAI 2025 · similarity 0.86)
- Accurate Scene Text Recognition with Efficient Model Scaling and Cloze Self-Distillation (CVPR 2025 · similarity 0.86)
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.
