1. Clone this repository:

   git clone https://github.com/zhayanping/MagicFuse.git cd MagicFuse

2. Create a Conda environment (recommended):

   conda create -n MagicFuse python=3.12 conda activate MagicFuse

3. Install dependency packages:

   pip install -r requirements.txt

Our project adopts a distributed training mode, you can modify the relevant settings in the train.py file to specify the appropriate CUDA device identifier for training. Please store the training data in the following format:

⚡ quicktest.ipynb Used for quick testing on single images. You can modify the input/output paths directly in the Jupyter Notebook to easily check the inference results of a single image.

⚡ test.py Designed for large-scale image testing tasks. This script supports multi-GPU parallel testing to efficiently process large batches of images. If you need to adjust the image size for testing, you can configure it in the dataset.py file.

⚡ test4largeImg.ipynb Specifically developed for testing large-size images. It adopts a dynamic model loading strategy to effectively save GPU memory usage.

If this work is helpful to you, please cite it as: