choidaedae / ddpm-scd

Semantic Change Detection Using Denoising Diffusion Probabilistic Model

Semantic Change Detection Using Denoising Diffusion Probabilistic Model (DDPM-scd)

• This work is done during the internship in Meissa Planet.
• This model is large vision model which can do some remote sensing image analysis tasks, such as change detection, semantic segmentation, and semantic change detection.

How to use it?

• It requires Docker, Pytorch 2.0.1, CUDA 11.8, CUDNN 8700.
• To make environment to get ready for running model, build Docker following below
• Or you can also use conda environment

if you want to use Docker

1. Docker build

  docker build {image_name} .

2. Docker run

  docker run --gpus all

If you want to use conda

1. Create Conda virtual environment

  conda create -n {environment_name} python=3.8.10

2. Activate Conda environment

  conda activate {environment_name}

3. Install all requirements to run model

  pip install -r requirements.txt

DDPM-SCD train script

1. If you want to training DDPM,

  python train_ddpm.py

2. If you want to training Pixel Classifier(Semantic Segmentation)

• It needs to pre-trained DDPM weight.

  python train_pc.py 

3. If you want to training CD Network

• It needs to pre-trained DDPM weight.

python train_cd.py

4. If you want to test your DDPM (Sampling Images),

python train_ddpm.py --config config/DDPM/ddpm_sampling.json --phase val

5. If you want to test your Pixel Classifier(Semantic Segmentation)

python test_pc.py --config config/PC/PC_{dataset_name}.json 

6. If you want to test your CD Network

python test_cd.py --config config/CD/{dataset_name}.json --phase test -log_eval

7. If you want to train your whole model

python train_model.py

8. If you want to test your whole model

python test_model.py