Is it possible to achieve accuracy = 99%?

Hi everyone!
As described in the assignment, we have the following constraint:
“Note that you should use no more than 2 Conv2D and 2 MaxPooling2D layers to achieve the desired performance.”
And the desired performance is:
“A reasonable benchmark is to achieve over 99% accuracy for training and over 95% accuracy for validation within 15 epochs.”
Has anyone managed to reach this level?
i’m stuck at about 90%. Any hints?

Hi @Maxim_Titarenko !
I’m glad to see you here.
what techniques did use to impove it ?

Hello Maxim! I struggled with it a bit, but I have managed to achieve:
→ loss: 0.0350 - accuracy: 0.9902 - val_loss: 0.0152 - val_accuracy: 0.9927

Did it with only 2 convolution/maxpooling layers and within 15 epochs. The answer is a bit counter-intuitive so keep at it. It’s worth it!

1 Like

can you please send your code?

Hi @Bosinceanu_Vladimir , is there any dropout in your model? trying to achieve 99%. Thanks ahead.

Hi fellow students,
After getting hints from this link and patiently tweaking number of filters on Conv2D, finally get it. Hope this is helpful for you.