SEC IPO Filing Images: 76,000+ Labeled Visuals and Charts (1994–2026)

76,000+ images extracted from U.S. SEC IPO registration statements (S-1 and F-1 filings) spanning 1994 to 2026. Created by gtfintechlab, the data was processed using a YOLOv8 detection pipeline and verified by an ensemble of eight Vision-Language Models.

Use Cases

• Training object detection models to locate charts in financial documents using YOLOv8 labels
• Analyzing the evolution of financial data visualization using the 'chart type' metadata
• Evaluating VLM performance on domain-specific financial imagery using the ensemble-verified classifications

Strengths

• 76,000+ labeled images
• Verification by an ensemble of 8 Vision-Language Models
• Covers a 30-year span of U.S. IPO filings

Limitations

• Automated classification pipeline may introduce label noise compared to human-expert annotation
• Limited to U.S. SEC filings (S-1 and F-1), excluding other international markets

Provenance

Source

SEC EDGAR system

Collection Method

Extracted from filings using YOLOv8 and annotated via an ensemble of 8 Vision-Language Models

Time Range

1994-2026

Freshness

Last updated February 2026.

Geography

United States