Skip to content

oier-namespace-std/2023MLproject

Repository files navigation

预处理图像

python main_preproc.py \
--data_dir \
/data/whma/Abdomen/RawData/Training \
--checkpoint \
sam_vit_h_4b8939.pth \
--device \
cuda \
--model_type \
vit_h

训练

python main_train.py \
--data_dir \
/data/whma/Abdomen/RawData/Training \
--checkpoint \
sam_vit_h_4b8939.pth \
--device \
cuda \
--model_type \
vit_h \
--loss_fn \ # 损失函数类别,参见utils.py
DCL \
--num_epochs \
20 \
--batch_size \
16 \
--num_workers \
4 \
--lr \
1e-5 \
--weight_decay \
5e-5 \
--save_dir \
model \
--prompter \ # 提示器类别
 single \
--train_class \ # 设置此项表示训练分类器
--train_prompt \ # 设置此项表示训练prompt_encoder
--grid # 设置此项表示使用为grid分割准备的训练

测试

python main_eval.py \
--data_dir \
/data/whma/Abdomen/RawData/Training \
--checkpoint \
sam_vit_h_4b8939.pth \
--device \
cuda \
--model_type \
vit_h \
--batch_size \
16 \
--num_workers \
4 \
--prompter \ # 提示器类别
 single \
# --train_class # 设置此项表示测试分类器

grid测试

python main_eval_grid.py \
--data_dir \
/data/whma/Abdomen/RawData/Training \
--checkpoint \
model/20230618-184844/sam_vit_h_ft.pth \
--device \
cuda \
--model_type \
vit_h \
--batch_size \
16

模型地址:https://disk.pku.edu.cn:443/link/8BD35BD9C247EB5FB2186078D85D58B0

About

PKU EECS Machine Learning Project

Resources

License

Code of conduct

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published