ICDAR 2026 Competition on Information Extraction from Atomic Layer Deposition/Etching (ALD/E) Scientific Figures
The benchmark shows multimodal models still falter when figures require extracting data or doing scientific reasoning.
The paper describes Sci-ImageMiner, an expert-annotated benchmark and ICDAR 2026 competition for ALD/E scientific figures. The competition ran from 9 January to 8 April 2026 and drew 68 active participants with 1,263 public and private submissions. Models performed better on classification and summarization than on data extraction and visual question-answering. HF Daily Papers' note
The paper describes Sci-ImageMiner, an expert-annotated benchmark and ICDAR 2026 competition for ALD/E scientific figures. The competition ran from 9 January to 8 April 2026 and drew 68 active participants with 1,263 public and private submissions. Models performed better on classification and summarization than on data extraction and visual question-answering. HF Daily Papers' note
score 4