Hi,
I was trying different things to achieve accuracy more than 99%.
One more conv layer, different optimizers with different learning rate accuracy was between 70-80%.
Just ImageDataGenerator with rescale, rotation and fill mode helped to achieve accuracy more than 99%.
Do we have any others approaches?
Best regards, Taras
can you pleasee send your code?
Hi, Taras.
“A reasonable benchmark is to achieve over 99% accuracy for training and over 95% accuracy for validation within 15 epochs.” This seems too high for me because the task is 26 class classification.
I checked more sample images than 100.
I think shear and rotation are useless or limited gain since all images I checked are upright.
And horizontal flips will be harmful because they are all right hand.
I reflect those findings on the settings of ImageDataGenerator but my result was far beyond “A reasonable benchmark”
Any tips on how to achieve 99% accuracy in 15 epochs? Did you use transfer learning?
I was running the calculations as I wrote my earlier post, and soon after that, I went to 99% in 15 epochs. It seemed better to be conservative with augmentation.
Thanks.
“It seemed better to be conservative with augmentation.” Of Course!!! This kind of data must not have too many variations.
hello shiro, your post helped me understand and cleared week 4 assignment. thank you. I understood one think the more number of classes, the lesser complicated should be one’s model.
I can’t even get higher than 4% accuracy and I’ve been following the advice listed here. I don’t know what to do at this point.