search

SaiHitesh16 / Glaucoma-Detection-using-CNN

Glaucoma detection using deep learning(cnn)
https://glaucoma-detector.streamlit.app/
MIT License
8 stars 3 forks source link

validation accuracy #2

Closed roshanshaw3117 closed 10 months ago

roshanshaw3117 commented 1 year ago

bro validation accuracy not coming as expected is there any solution of it?

SaiHitesh16 commented 1 year ago

are u trying on the same datasets?

roshanshaw3117 commented 1 year ago

yes validation accuracy is around 70%

SaiHitesh16 commented 1 year ago

Can you provide the screenshot of accuracy graph or epochs so that I may help you

roshanshaw3117 commented 1 year ago

Epoch 148/150 32/32 [==============================] - 32s 914ms/step - loss: 0.0019 - accuracy: 1.0000 - val_loss: 1.1507 - val_accuracy: 0.6797 Epoch 149/150 32/32 [==============================] - 31s 991ms/step - loss: 8.3529e-04 - accuracy: 1.0000 - val_loss: 0.9458 - val_accuracy: 0.6562 Epoch 150/150 32/32 [==============================] - 32s 1s/step - loss: 0.0020 - accuracy: 1.0000 - val_loss: 1.0846 - val_accuracy: 0.6797

SaiHitesh16 commented 1 year ago

The model was overfitted. How many images are you using in total and for training,validation.

roshanshaw3117 commented 1 year ago

same as yours do i need to ensemble

SaiHitesh16 commented 1 year ago

I have not used it. Try to re run the epochs code and check whether there is any change and also try to decrease the epochs value because training accuracy is 1.00.

what is test accuracy ?

roshanshaw3117 commented 1 year ago

65%

SaiHitesh16 commented 1 year ago

ok try changing split ratio and check or can you provide the link for your code so that I can understand the issue better.

roshanshaw3117 commented 1 year ago

ok