Published October 27, 2025 | Version v1

Graph and Image Representations of Vulnerable and Non-Vulnerable Software Functions

Authors/Creators

Description

The dataset contains image representations of source code files converted into three types of program graph-based views:
AST
PDG
CFG

The root directory includes three folders corresponding to these graph types:
dataset/
  ├── ast/
  ├── pdg/
  └── cfg/

Inside each graph-type folder, the images are categorized into two classes based on vulnerability status:
ast/
  ├── Vul/      # Vulnerable instances
  └── No-Vul/   # Non-vulnerable instances

pdg/
  ├── Vul/
  └── No-Vul/

cfg/
  ├── Vul/
  └── No-Vul/

The total number of images in each class across all graph types is:
No-Vul: 21,031 instances
Vul: 12,294 instances
All images are stored in .pkl format, where each file contains the serialized pixel data of a single graph image.

Files

dataset.zip

Files (2.5 GB)

Name Size
md5:27b2d8b1369123ad3f44cbd77d24e6e4
2.5 GB Preview Download

Additional details

Related works

Cites
Conference proceeding: 10.1145/3510003.3510229 (DOI)

Dates

Submitted
2025-10-27

References

  • VulCNN: an image-inspired scalable vulnerability detection system