Image Defect Generation Service

This repository provides an example of how to deploy a Stable Diffusion based image defect generator on CentOS 7. The service exposes a simple HTTP API that accepts an input image and a defect type (0, 1 or 2) and returns an image with the requested defect applied. If the uploaded image matches one of the pre-generated samples (checked via MD5), the server returns the pre-generated result instead of creating a new image.

The repository contains:

• requirements.txt – Python package requirements.
• server.py – FastAPI service that runs the model and performs MD5 checking.
• dataset_md5.py – helper script to build a JSON mapping of MD5 hashes to existing defect images.
• package_env.sh – optional script showing how to download all required Python wheels for offline installation.

1. Preparing CentOS 7

1. Install system packages:

   sudo yum install -y epel-release
   sudo yum install -y git python3 python3-venv

2. (Optional) Install NVIDIA drivers and CUDA if GPU acceleration is needed. Refer to the official NVIDIA documentation for CentOS 7.

2. Create Python Environment

1. Clone this repository and enter it:

   git clone <repo-url>
   cd Defect

2. Create and activate a virtual environment:

   python3 -m venv venv
   source venv/bin/activate

3. Install the required Python packages:

   pip install -r requirements.txt

If running on a machine without internet access, execute package_env.sh on a machine that does have internet. It will download all required wheels into wheels/. Copy that directory to the target machine and install with pip install wheels/*.

3. Prepare the MD5 Mapping

Use dataset_md5.py to create a JSON file containing MD5 hashes of the normal images mapped to the corresponding defect images. Example:

python dataset_md5.py data/md5_map.json \
  --pairs "/path/to/设备破损/ref" "/path/to/设备破损/def" \
  --pairs "/path/to/设备烟火/ref" "/path/to/设备烟火/def" \
  --pairs "/path/to/渗漏油/ref" "/path/to/渗漏油/def"

The resulting data/md5_map.json will be loaded by server.py at runtime.

4. Running the API Service

Start the API with:

uvicorn server:app --host 0.0.0.0 --port 8000

Request format:

• URL: /generate
• Method: POST
• Form fields:
  • defect_type – integer 0, 1 or 2.
  • file – image file to process.

Example using curl:

curl -F defect_type=1 -F file=@input.jpg http://localhost:8000/generate --output result.png

The service will respond with the defect image. If the uploaded image matches an entry in md5_map.json, the corresponding pre-generated image is returned. Otherwise the model generates a new image using Stable Diffusion.

5. Packaging for Deployment on Multiple Servers

1. On a machine with internet access, run package_env.sh. This downloads all Python wheels specified in requirements.txt into the wheels/ directory.
2. Copy the repository and the wheels/ directory to the target server.
3. Create a virtual environment and install the wheels:

   python3 -m venv venv
   source venv/bin/activate
   pip install wheels/*

4. Copy or recreate the data/md5_map.json file on each server.
5. Start the API as shown above.

With this approach you can replicate the same environment on multiple CentOS 7 servers without re-downloading packages from the internet.