Datasets:
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Error code: JobManagerCrashedError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
image
image |
|---|
SEC IPO Filing Image Dataset
A large-scale, labeled dataset of 76,000+ images extracted from U.S. IPO registration statements (S-1 and F-1 filings) filed with the SEC EDGAR system, spanning 1994–2026.
Every image has been classified through a multi-stage pipeline: initial detection with YOLOv8, followed by verification from an ensemble of 8 Vision-Language Models (VLMs). Chart images include additional structured metadata describing chart type, visual properties, and content.
Classification Pipeline
Labels
Each image is assigned one of the following verified labels:
| Label | Description |
|---|---|
chart |
Bar charts, line charts, pie charts, organizational charts, flow charts, etc. |
logo |
Company logos and branding marks |
map |
Geographic maps |
infographic |
Composite visual displays combining data, icons, and text |
other |
Decorative images, photographs, signatures, and other non-classifiable visuals |
Fields
Image & Classification
| Field | Type | Description |
|---|---|---|
file_name |
string |
Relative path to the image file |
label |
string |
Final verified image class (chart, logo, map, infographic, other) |
initial_yolo_prediction |
string |
Raw classification from the YOLOv8 object detection model |
llm_yolo_verification_score |
float |
Consensus score (0.0–1.0) across the VLM ensemble verifying the YOLO prediction. null when the initial prediction was other and no verification was needed. |
llm_yolo_verification_votes |
object |
Per-model votes (true/false) from each of the 8 VLMs used for verification |
image_width |
int |
Image width in pixels |
image_height |
int |
Image height in pixels |
Chart Details (charts only)
| Field | Type | Description |
|---|---|---|
llm_chart_details |
object |
Structured attributes for chart images (e.g., chart types, is_3d, has_legend, axis labels). null for non-chart images. |
llm_chart_details_votes |
object |
Per-model votes for the chart detail extraction. null for non-chart images. |
Company & Filing Metadata
| Field | Type | Description |
|---|---|---|
company_name |
string |
Name of the company filing the IPO |
company_tickers |
list[string] |
Ticker symbol(s), if available at time of filing |
cik |
string |
SEC Central Index Key — unique identifier for the filing entity |
sic |
string |
Standard Industrial Classification code |
industry |
string |
Human-readable industry description derived from the SIC code |
office |
string |
SEC office responsible for reviewing the filing. null if unavailable. |
exchanges |
list[string] |
Stock exchange(s) the company listed on, if available |
filing_type |
string |
SEC form type (e.g., S-1, S-1/A, F-1, F-1/A) |
filing_date |
string |
Date the filing was submitted (YYYY-MM-DD) |
filing_accession_number |
string |
Unique SEC accession number for the filing |
Source URLs
| Field | Type | Description |
|---|---|---|
image_url |
string |
Direct URL to the image on SEC EDGAR |
filing_url |
string |
URL to the source filing document on SEC EDGAR |
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}
}
- Downloads last month
- 26