The 1st Sci-ImageMiner Competition: Information Extraction from Scientific Figures in Materials Science

Focus: Quantitative plots from Atomic Layer Deposition and Etching (ALD/E) research

Organized as part of the ICDAR 2026 Competition track

ICDAR 2026 — The 20th International Conference on Document Analysis and Recognition 30 Aug – 04 Sep 2026 in Vienna, Austria

Motivation

Scientific figures often contain critical results that never appear explicitly in the text. Despite recent advances in multimodal large language models, most existing benchmarks rely on generic or synthetic visuals. Sci-ImageMiner addresses this gap by introducing a curated benchmark grounded in authentic scientific figures from a specialized scientific domain.

The following tasks are available:

• Task 1. Figure Classification – identify the chart or figure type
• Task 2. Data Table Extraction – reconstruct the underlying tabular data from quantitative plots
• Task 3. Figure Summarization – generate concise, factual summaries of key trends
• Task 4. Visual Question Answering (VQA) – answer scientific questions requiring reasoning over figure content

Teams may participate in any subset of tasks (including all four).

Sci-ImageMiner Competition Website: https://sites.google.com/view/sci-imageminer/