Research map / Detection and segmentation / CNN features
COO: Comic Onomatopoeia Dataset for Recognizing Arbitrary or Truncated Texts
Jeonghun Baek, Yusuke Matsui, Kiyoharu Aizawa
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 |
|---|---|---|---|
| train with polygon instead of bounding box annotation (ablation) | ABCNet v2 · COO text detection (test) | Hmean: 61.2 -> 66.0 | Improved |
| train with polygon instead of bounding box annotation (ablation) | MTS v3 · COO text detection (test) | Hmean: 62.5 -> 67.8 | Improved |
| use segmentation-based MTS v3 instead of regression-based ABCNet v2 | text detector · COO text detection (test) | Hmean: 66.0 -> 67.8 | Improved |
| add rotation trick (ablation) | TRBA · COO text recognition (test) | word accuracy: 46.3 -> 49.8 | Improved |
| add SAR decoding alone (ablation) | TRBA · COO text recognition (test) | word accuracy: 46.3 -> 43.2 | Hurt |
| add SAR decoding on top of rotation trick (ablation) | TRBA · COO text recognition (test) | word accuracy: 49.8 -> 55.4 | Improved |
| add Hard RoI masking (ablation) | TRBA + rotation + SAR · COO text recognition (test) | word accuracy: 55.4 -> 63.5 | Improved |
| fill half of each mini-batch with original images with Hard RoI (ablation) | TRBA + rotation + SAR · COO text recognition (test) | word accuracy: 63.5 -> 67.9 | Improved |
| add 2D attention (input height 64) (ablation) | TRBA · COO text recognition (test) | word accuracy: 67.9 -> 78.5 | Improved |
| add 2D attention (input height 100) (ablation) | TRBA · COO text recognition (test) | word accuracy: 67.9 -> 81.0 | Improved |
| use M4C-COO instead of distance rule baseline | link prediction · COO link prediction (test) | Hmean: 2.1 -> 72.7 | Improved |
| restrict to only five vocabularies instead of 11,640 (ablation) | M4C-COO · COO link prediction (test) | Hmean: 45.4 -> 72.7 | Improved |
Closest papers by meaning
- The Manga Whisperer: Automatically Generating Transcriptions for Comics (CVPR 2024 · similarity 0.83)
- ASCIIEval: Benchmarking Models' Visual Perception in Text Strings via ASCII Art (ICLR 2026 · similarity 0.81)
- Advancing Manga Analysis: Comprehensive Segmentation Annotations for the Manga109 Dataset (CVPR 2025 · similarity 0.80)
- BBScore: A Brownian Bridge Based Metric for Assessing Text Coherence (AAAI 2024 · similarity 0.80)
- Alchemist: Turning Public Text-to-Image Data into Generative Gold (NeurIPS 2025 · similarity 0.79)
- Cracking the Code of Juxtaposition: Can AI Models Understand the Humorous Contradictions (NeurIPS 2024 · similarity 0.79)
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.
