train output loss diffurent from the test output loss

[Usernamezhx]

hi
At first . thanks for your code . but I have a problem when i train with the ResNet_50_train_val.prototxt . Iteration 2w. Train output loss diffurent from the test output loss. the test loss was 2.0+ . what i have changed is the batch_size : I change it to 28 because of the memoray. others were same to your ResNet_50_train_val.prototxt and solver.prototxt . hope for your replying.

[antingshen]

Not sure I understand what your problem is

[Usernamezhx]

thanks for your reply . i hope you can waste several minuts to read the log file .it is not long .
my_log.txt
i set the Test batch_size : 1 in train_val.prototxt . so if i want to test all my test sample . i have to set the test_iter: 100000. because my test sample is 100000 .it is so big that i will take 1 hour to finish it.
it is wrong ?
hope your reply

[antingshen]

Your train loss isn't going down, so your net isn't learning. I'd hold on the testing until training is working.

I can't help you figure out why your net isn't learning, but one thing to look at is to make sure that you increased your iter_size when decreasing batch size to ensure the same mathematical batch size for updates.

[qdinfish]

@Usernamezhx Hi, I encounter the same issue, have you fund the root cause ?

[ghost]

@Usernamezhx your training loss is way too small here, I'd guess that your training accuracy is pretty high while testing is somewhat lower than expected. Are you using ImageNet here? One of the issues could be the dataset leading to the network overfitting.

Were you able to solve this issue?