Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Loss not changing when training with Kaggle dataset #1

Open
Hamdi-Ben-Abdallah opened this issue Mar 3, 2022 · 3 comments
Open

Loss not changing when training with Kaggle dataset #1

Hamdi-Ben-Abdallah opened this issue Mar 3, 2022 · 3 comments

Comments

@Hamdi-Ben-Abdallah
Copy link

Hello,

Thank you for this beautiful work

I have not changed the settings and am using your code as it is with the Kaggle dataset but I have the loss that does not change when training.

Why don't I get the same results as you?
Thank you in advance.

Epoch 0 | Loss 0.45
Epoch 1 | Loss 0.42
Epoch 2 | Loss 0.42
Epoch 3 | Loss 0.42
Epoch 4 | Loss 0.41
Epoch 5 | Loss 0.41
Epoch 6 | Loss 0.41
Epoch 7 | Loss 0.41
Epoch 8 | Loss 0.41
Epoch 9 | Loss 0.41
Epoch 10 | Loss 0.41
Epoch 11 | Loss 0.41
Epoch 12 | Loss 0.41
Epoch 13 | Loss 0.41
Epoch 14 | Loss 0.41
Epoch 15 | Loss 0.41
Epoch 16 | Loss 0.41
Epoch 17 | Loss 0.41
Epoch 18 | Loss 0.41
Epoch 19 | Loss 0.41
Epoch 20 | Loss 0.41
Epoch 21 | Loss 0.41
Epoch 22 | Loss 0.41
Epoch 23 | Loss 0.41
Epoch 24 | Loss 0.41
Epoch 25 | Loss 0.41
Epoch 26 | Loss 0.41
Epoch 27 | Loss 0.41
Epoch 28 | Loss 0.41
Epoch 29 | Loss 0.41
Epoch 30 | Loss 0.41
Epoch 31 | Loss 0.41
Epoch 32 | Loss 0.41
Epoch 33 | Loss 0.41
Epoch 34 | Loss 0.41
Epoch 35 | Loss 0.41
Epoch 36 | Loss 0.41
Epoch 37 | Loss 0.41
Epoch 38 | Loss 0.41
Epoch 39 | Loss 0.41
Epoch 40 | Loss 0.41
Epoch 41 | Loss 0.41
Epoch 42 | Loss 0.41
Epoch 43 | Loss 0.41
Epoch 44 | Loss 0.41
Epoch 45 | Loss 0.41
Epoch 46 | Loss 0.41
Epoch 47 | Loss 0.41
Epoch 48 | Loss 0.41
Epoch 49 | Loss 0.41
Train Accuracy = 0.8181365664610407

@Utkarsh-Deshmukh
Copy link
Owner

can you try re-training once only with de-focus blurred and sharp images (not using the motion blur images)
If you see that the training accuracy is high, then you can think of introducing motion blur images to the training.

@Hamdi-Ben-Abdallah
Copy link
Author

I re-trained the model with blurred and sharp images without using motion blur images but the accuracy of the training is very low.
I have an accuracy of about 40% with the same setup as you. The only change I made is :
batch_size = 32
num_epochs = 1000

@Utkarsh-Deshmukh
Copy link
Owner

can you try to run the pre-trained model on the test data using the script Test_main.py and see what kind of performance you get?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants