This is the official repo of "The Crack Diffusion Model: An Innovative Diffusion-Based Method for Pavement Crack Detection".

Suggest train and test on Linux.

The command for training:

On Single GPU

CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.run --nproc_per_node=1 multi_task_train.py

On Multiple GPU

CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.run --nproc_per_node=2 multi_task_train.py

Change from multi_task_train.py to multi_task_test.py for testing.

If train and test on Windows,

remove notation # os.environ["CUDA_VISIBLE_DEVICES"]='0' # if train on windows in multi_task_train.py and # args.dist_backend = 'gloo' # if train on windows in distribute_utils.py.

Use

python -m torch.distributed.run --nproc_per_node=1 multi_task_train.py