See corresponding paper accepted for 2nd Workshop on Intelligent Infocommunication Networks, Systems and Services (WINS 2024).
Train and test semantic segmentation guided diffusion models.
You will need to install: pytorch-lightning, torchvision, hydra, wandb, omegaconf (and skimage for BDD preprocessing)
Use: pip install -r requirements.txt
The data needs to be in a specific format. Our experiments utilized the Berkeley Deep Drive Dataset. To see detailed steps for that specific dataset, see data/BDD_instructions.md.
For custom datasets, you should have a fromat presented in data/sample . Two separate folders: one for the images, one for the masks and a colormap.json file, that contains mappings for the given classes.
You need to edit config/base.yaml with the correct paths. Make sure to set correct class number and image size as well.
Make sure you have configured everything in config/base.yaml !
Then use python train.py.
Make sure you have configured everything in config/base.yaml !
Then use python test.py.