This repository hosts the dataset resources for training and benchmarking image editing models.
All paths are relative to the repository root when cloned via Git.
📂 Dataset Structure
After cloning, you should have the following directories:
project_name/
├── dataset/
│ └── image_edit/
│ └── OmniEdit/ # Training data
└── edit_benchmark/
└── data/ # Benchmark data
Training Data
The training set is located under:
dataset/image_edit/OmniEdit
This directory contains images we need.
Benchmark Data
The benchmark set is located under:
edit_benchmark/data
This directory provides standardized test instructions and images used for evaluating editing performance across multiple metrics.
📑 Notes
All paths are relative to the repository root after cloning.
Simply download or clone the repository and place the files into the corresponding directories shown above.
The benchmark is designed to be consistent with common editing evaluation setups (e.g., GEdit, ImgEdit-Bench).