ALD/E-ImageMiner Benchmark Targets Scientific Image Comprehension
The newly introduced benchmark, ALD/E-ImageMiner, seeks to enhance the capacity of multimodal AI systems in retrieving and interpreting scientific tables and figures. Detailed in arXiv:2608.14075, it comprises 1,951 figures sourced from 205 publications, all expertly annotated for tasks such as classification, data table extraction, summarization, and visual question answering. This benchmark will also serve as the foundation for the ICDAR 2026 Competition focused on Information Extraction from Atomic Layer Deposition/Etching Scientific Figures. The accompanying proceedings offer insights on how this benchmark could shape future challenges in scientific imaging. The authors highlight 'scientific conceptual understanding from images' as a long-term goal, emphasizing its importance for digital libraries and AI systems that face difficulties with scientific figures.
Key facts
- The benchmark is called ALD/E-ImageMiner.
- It provides 1,951 figures from 205 publications.
- Figures are expert-annotated for classification, data table extraction, summarization, and visual question answering.
- It is the basis for the ICDAR 2026 Competition on Information Extraction from Atomic Layer Deposition/Etching Scientific Figures.
- The companion proceedings are published on arXiv with ID 2608.14075.
- The benchmark aims to improve multimodal AI systems' retrieval and interpretation of scientific figures.
- The long-term objective is 'scientific conceptual understanding from images'.
- The work addresses the difficulty digital libraries and AI systems face with scientific figures.
Entities
Institutions
- arXiv
- ICDAR