search

ECCV2024MSE / ECCV_MSE_Teacher

1 stars 0 forks source link

Choose of loss function #1

Open OrangeSodahub opened 3 months ago

OrangeSodahub commented 3 months ago

@deponce Hi, for this line: https://github.com/ECCV2024MSE/ECCV_MSE_Teacher/blob/76f6c300e45babc1e15188a26ed9e7624b34c25c/train_teacher.py#L109 you choose loss function between "MSE" and "CE" via opt.loss_func.upper, there is one thing I want to make it clear, did you use the exactly same code to conduct your experiments? Thanks

myl666666 commented 3 months ago

Hello, how should I train the student model? Why did I report this error when running the student model? Is the result of the experiment the test result of the student model? D:\anaconda\envs\qqq\python.exe E:\desktop\ECCV_MSE_Teacher-main\ECCV_MSE_Teacher-main\train_student.py Traceback (most recent call last): File "E:\desktop\ECCV_MSE_Teacher-main\ECCV_MSE_Teacher-main\train_student.py", line 354, in main() File "E:\desktop\ECCV_MSE_Teacher-main\ECCV_MSE_Teacher-main\train_student.py", line 153, in main opt = parse_option() File "E:\desktop\ECCV_MSE_Teacher-main\ECCV_MSE_Teacher-main\train_student.py", line 109, in parse_option opt.model_t, lr_t = get_teacher_name(opt.path_t) File "E:\desktop\ECCV_MSE_Teacher-main\ECCV_MSE_Teacher-main\train_student.py", line 132, in get_teacher_name segments = modelpath.split('/')[-2].split('') AttributeError: 'NoneType' object has no attribute 'split'

OrangeSodahub commented 3 months ago

@myl666666 I'm not the author, don't ask me...

myl666666 commented 2 months ago

@deponce Hi, for this line:

https://github.com/ECCV2024MSE/ECCV_MSE_Teacher/blob/76f6c300e45babc1e15188a26ed9e7624b34c25c/train_teacher.py#L109

you choose loss function between "MSE" and "CE" via , there is one thing I want to make it clear, did you use the exactly same code to conduct your experiments? Thanksopt.loss_func.upper

criterion = nn.MSELoss() if opt.loss_func.upper() == 'MSE' else nn.CrossEntropyLoss() ?