Research map / Detection and segmentation / Open-set and few-shot detection
CEDe: A collection of expert-curated datasets with atom-level entity annotations for Optical Chemical Structure Recognition
Rodrigo Hormazabal, Changyoung Park, Soonyoung Lee, Sehui Han, Yeonsik Jo, Jaewan Lee, Ahra Jo, Seung Hwan Kim, Jaegul Choo, Moontae Lee, Honglak Lee
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 synthetic pretraining (1M images) before real data (ablation) | CNN+GRU image-to-SMILES · CEDe UOB | graph matching accuracy (%): 5.12 -> 32.01 | Improved |
| add synthetic pretraining (10K images) before real data (ablation) | FasterRCNN+Instance-pair Transformer · CEDe UOB | graph matching accuracy (%): 49.77 -> 67.94 | Improved |
| fine-tune on 20% of real data after synthetic pretraining (ablation) | FasterRCNN+Instance-pair Transformer · CEDe UOB | graph matching accuracy (%): 67.94 -> 89.26 | Improved |
| fine-tune on 20% of real data after synthetic pretraining (ablation) | CNN+GRU · CEDe JPO | graph matching accuracy (%): 15.28 -> 33.61 | Improved |
| fine-tune on real data after synthetic pretraining (ablation) | CNN+GRU · CEDe UOB | graph matching accuracy (%): 32.01 -> 62.57 | Improved |
| use chemical entity detection with 10K data instead of CNN+GRU with 1M data | FasterRCNN+bond bbox · CEDe synthetic test | graph matching accuracy (%): 61.2 -> 85.92 | Improved |
| use FasterRCNN+bond bbox detection instead of rule-based Imago | FasterRCNN+bond bbox · CEDe JPO fine-tuned | graph matching accuracy (%): 41.39 -> 74.12 | Improved |
| use FasterRCNN+IPT instead of OSRA | FasterRCNN+Instance-pair Transformer · CEDe CLEF fine-tuned | graph matching accuracy (%): 92.84 -> 81.64 | Hurt |
| use image-to-SMILES CNN+Transformer instead of detection with synthetic pretraining | CNN+Transformer · CEDe USPTO synthetic only | graph matching accuracy (%): 75.43 -> 42.60 | Hurt |
| fine-tune CNN+GRU+Attention on real data (ablation) | CNN+GRU+Attention · CEDe CLEF | graph matching accuracy (%): 23.18 -> 39.19 | Improved |
| fine-tune CNN+GRU+Attention on real data (ablation) | CNN+GRU+Attention · CEDe JPO | graph matching accuracy (%): 17.22 -> 30.28 | Improved |
| fine-tune FasterRCNN+bond bbox on real data (ablation) | FasterRCNN+bond bbox · CEDe JPO | graph matching accuracy (%): 49.44 -> 74.12 | Improved |
Closest papers by meaning
- Atom-Level Optical Chemical Structure Recognition with Limited Supervision (CVPR 2024 · similarity 0.93)
- MarkushGrapher: Joint Visual and Textual Recognition of Markush Structures (CVPR 2025 · similarity 0.90)
- RFL: Simplifying Chemical Structure Recognition with Ring-Free Language (AAAI 2025 · similarity 0.89)
- ChemVLM: Exploring the Power of Multimodal Large Language Models in Chemistry Area (AAAI 2025 · similarity 0.86)
- TinyChemVL: Advancing Chemical Vision-Language Models via Efficient Visual Token Reduction and Complex Reaction Tasks (AAAI 2026 · similarity 0.86)
- ChemPile: A 250 GB Diverse and Curated Dataset for Chemical Foundation Models (NeurIPS 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.
