Capstone Figures: Scientific Images from ICLR, NeurIPS, and ICML Conferences

Scientific figure images extracted from papers published at ICLR 2024-2025, NeurIPS 2023-2024, and ICML 2023-2024 conferences. The dataset is intended for multidimensional figure quality assessment and was created by Jhao0930. It was last updated on May 21, 2026.

Use Cases

• Train image classification models to categorize figure types based on the visual content from major AI conferences.
• Develop models for figure quality assessment based on the multidimensional quality annotations mentioned in the description.
• Analyze visual trends in scientific publications from ICLR, NeurIPS, and ICML based on the extracted figure images.
• Benchmark image captioning or description generation models on scientific figures.

Strengths

• Images are sourced from recent, high-profile AI/ML conferences (ICLR, NeurIPS, ICML) spanning 2023-2025.
• Includes structured JSONL annotation files for figure quality assessment, as mentioned in the companion GitHub repo.
• File naming convention includes venue, year, and figure number, which likely provides traceability to source papers.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale model training.
• The license is unspecified, which may restrict commercial or redistribution use.

Provenance

Source

Papers from ICLR, NeurIPS, and ICML conferences.

Collection Method

Extracted from conference papers, likely via automated scripts.

Time Range

2023-2025

Freshness

Last updated 2026-05-21 10:40:15; freshness should be verified.