COVID-CT-Similarity-Regularization

Train SR models with scaling schedulers

Constant Scheduler

python -m torch.distributed.launch --nproc_per_node 4 --master_port 12345 main_constraint_level.py --model-arch densenet121 --batch-size 64 --aug-level 4 --tag constant_l4

Linear Scheduler

python -m torch.distributed.launch --nproc_per_node 4 --master_port 12345 main_constraint_level.py --cfg configs/scaling_linear.yaml --model-arch densenet121 --batch-size 64 --aug-level 4 --tag linear_l4

Cosine Scheduler

python -m torch.distributed.launch --nproc_per_node 4 --master_port 12345 main_constraint_level.py --cfg configs/scaling_cosine.yaml --model-arch densenet121 --batch-size 64 --aug-level 4 --tag cosine_l4

Train original models

python -m torch.distributed.launch --nproc_per_node 4 --master_port 12345 main_level.py --model-arch densenet121 --batch-size 64 --aug-level 4 --tag original_l4

Conventional BYOL pretrain + finetunning

Pretrain

python -m torch.distributed.launch --nproc_per_node 4 --master_port 12345 main_SSL.py
--batch-size 64 --model-arch densenet121 --aug-level 4 --tag byol_l4

Linear Evaluation

python -m torch.distributed.launch --nproc_per_node 4 --master_port 12345 main_SSL_linear_eval.py
--batch-size 256 --model-arch densenet121 --aug-level 4 --tag linear_BYOL_4

Resume from pretrained models

python -m torch.distributed.launch --nproc_per_node 4 --master_port 12345 main.py
--batch-size 64 --model-arch densenet121 --aug-level 4 --tag l4_BYOL_pretrain
--resume output/SSL/densenet121/byol_l4/ckpt_densenet121.pth