gtfintechlab/ipo-images

SEC IPO (S-1/F-1) Image Dataset

Dataset Description

This is a comprehensive dataset of images extracted from U.S. IPO Registration Statements (S-1 and F-1 filings) filed with the SEC between 1994 and 2026.

The dataset focuses on image understanding and chart analysis. Each image has been classified and verified using a multi-stage pipeline involving object detection (YOLOv8) and an ensemble of Vision Language Models (VLMs).

Data Structure

The dataset uses a JSONL metadata file mapping to the raw images. Key fields include:

• image: The pixel data of the extracted image.
• label: The verified class of the image (Chart, Logo, Map, Infographic, Other).
• initial_yolo_prediction: The raw classification from the object detection model.
• llm_yolo_verification_score: A confidence score representing the consensus of VLMs verifying the initial prediction.
• llm_chart_details: Detailed attributes for images labeled as 'chart' (e.g., is_3d, has_legend, chart_types).
• company_name: The name of the company filing the IPO.
• cik: A unique identifier for the company filing the IPO.
• sic A value used to identify which industry the company filing the IPO operates in.
• filing_url: The source URL of the filing on the SEC EDGAR system.
• image_url: The source URL of the image on the SEC EDGAR system.

Usage Example

You can load this dataset directly using the Hugging Face datasets library.

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("gtfintechlab/ipo-images", split="train")

# Access an example
example = dataset[0]

# Print metadata
print(f"Company: {example['company_name']}")
print(f"Label: {example['label']}")
if example['label'] == 'chart':
    print(f"Details: {example['llm_chart_details']}")

# View the image
example['image'].show()

Citation

The dataset is a product of the following research. If you use this dataset in your work, please cite:

@misc{galarnyk2026ipomine,
  title  = {IPO-Mine: A Toolkit and Dataset for Section-Structured Analysis of Long, Multimodal IPO Documents},
  author = {Galarnyk, Michael and Lohani, Siddharth and Nandi, Sagnik and Patel, Aman and Kannan, Vidhyakshaya and Banerjee, Prasun and Routu, Rutwik and Ye, Liqin and Hiray, Arnav and Somani, Siddhartha and Chava, Sudheer},
  year   = {2026},
  url    = {[https://huggingface.co/datasets/gtfintechlab/ipo-images](https://huggingface.co/datasets/gtfintechlab/ipo-images)},
  note   = {Preprint/Working Paper}
}