SAMDiffusion

SAMDiffusion: Semantic Segmentation with Diffusion Model and Segmentation Anything Model

Conda env installation

conda create -n SAMDiffusion python=3.8

conda activate SAMDiffusion

install pydensecrf https://github.com/lucasb-eyer/pydensecrf
pip install git+https://github.com/lucasb-eyer/pydensecrf.git

pip install -r requirements.txt

1. Prepare SAM2

Refer to facebookresearch/sam2 for the recommended folder structure.
Place the sam2 folder in the project directory, prepare the model files, and organize them as follows:

.
├── SAMDiffusion
│   ├── sam2_optimization.py
│   ├── sam2_optimization_multi.py
│   └── ...
└── sam2
    ├── checkpoints
    ├── configs
    └── ...

2. Data and mask generation

# generating data and attention map witn stable diffusion (Before generating the data, you need to modify the "hunggingface key" in the "Stable_Diffusion" codes to your own key. )
sh ./generate/VOC_multiple_data_generation.sh

3. Training model with clear data

Following the segmentation model framework from mmsegmentation,
we adopt the standard training configuration.

4. Our Synthetic Dataset

We are providing synthetic data here with Baidu Drive (password: abcd). Feel free to use it.