Bounding Box-Guided Diffusion for Industrial Image Synthesis

This repository contains the official implementation of the paper: "Bounding Box-Guided Diffusion for Synthesizing Industrial Images and Segmentation Map" accepted at Synthetic Data for Computer Vision Workshop - CVPR 2025

📌 Overview

This project introduces a diffusion-based generative framework guided by bounding boxes to synthesize high-quality industrial images along with corresponding segmentation maps. The method is designed to support precise localization, multi-part control, and mask generation, facilitating dataset creation for downstream tasks like defect detection and segmentation.

🖇️ Setup

Clone the repository and install the necessary dependencies:

git clone https://github.com/covisionlab/diffusion_labeling
cd diffusion_labeling
python3 -m venv .venv
source .venv/bin/activate   # On Windows use `.venv\Scripts\activate`
pip install -r requirements.txt

🚀 Usage

The repo is composed by three modules. That should be run consequentely:

1. preprocess: this module preprocess the original wood dataset which can be found here https://zenodo.org/records/4694695#.YkWqTX9Bzmg. Read the preprocess/README.md for more information.

2. generation: this module run the diffusion pipeline described in the paper, and generates the synthetic data which will be used in 3. Read the generation/README.md for more information.

3. segmentation: this is the segmentation module which should be run at the end of the pipeline to retrieve the metrics ebr, fid, sae, f1. Read the segmentation/README.md for more information.

We provide inside data/splits the official splits of the dataset we used to train our diffusion. So you can replicate the results.

📄 Citation

If you use this code in your research, please cite:

@inproceedings{
    simoni2025bounding,
    title={Bounding Box-Guided Diffusion for Synthesizing Industrial Images and Segmentation Maps},
    author={Alessandro Simoni and Francesco Pelosin},
    booktitle={Synthetic Data for Computer Vision Workshop @ CVPR 2025},
    year={2025}
}