Motivation

Scientific figures often contain the most important results of a paper—yet much of this information never appears in the text.

In materials science, and especially in Atomic Layer Deposition and Etching (ALD/E) research, quantitative plots such as XPS spectra, adsorption isotherms, phase diagrams, and band diagrams are central to scientific reasoning. These figures encode trends, values, and relationships that are difficult—or impossible—to recover from text alone.

Despite recent progress in multimodal large language models, most existing benchmarks rely on generic or synthetic images, leaving models poorly equipped to interpret authentic scientific figures from specialized domains.

Sci-ImageMiner addresses this gap by focusing on real ALD/E figures, starting with quantitative plots. The competition challenges participants to develop methods that can understand, interpret, and extract information from scientific figures as they appear in real research papers—supporting tasks such as classification, structured data extraction, summarization, and visual question answering.

By grounding multimodal information extraction in a real scientific domain, Sci-ImageMiner aims to advance figure understanding, domain adaptation, and scientific reasoning in document analysis and multimodal AI.

On the rest of this website, you will find detailed task descriptions, dataset access, evaluation protocols, and important dates. We invite you to explore the site and consider participating in the Sci-ImageMiner Challenge. 

Lead Organizers

Co-Organization Team

Funding Statement

Sci-ImageMiner is organized as an ICDAR 2026 competition and is supported by the NFDI4DataScience initiative (DFG, German Research Foundation, Grant ID: 460234259).