Accuracy In Fine Tuning

#1

I am getting very horrible accuracy of about 25% while Fine Tuning. Why I am getting this type to accuracy?

#2

Please Share your ipynb file using google drive.

#3

https://drive.google.com/open?id=1wQo1FEhwJ_gISi2ddepXFI3kSVpYN2yD

#4

I have used google colab

#5

Don’t need to normalize images in this case, I just removed division by 255.0 and got 95% validation accuracy in first 3-4 epochs. The reason behind not normalizing the image data is that, resnet50 model already contains Batch Normalization layer, so it automatically normalizes the outputs of previous layers, You can read more about batch normalization on blogs.

Hope this helped you :blush:

#6

That’s why i was getting very low Accuracy. Now i have removed that part and got good result. Thank you sir. :slight_smile: