@prashant_ml
I have created a model using mobilenet.
It is working fine for training and cross-validation data but after submitting results for test data it shows low accuracy.
Now the issue is with the labels… I have used generator function to load training and testing data. Now as we know that labels are automatically decided when we use generator function. Same thing happens here. For example Aerodactyl class should be label 0 but now it is label 4.
Is there any solution for the same or I need to remove generator function.
Here is my repo link. I haven’t pushed the predictions results though.
Test Accuracy is Low
hey @Joy-Gupta-2763139277246091 ,
I just checked your code , and its really good.
But , for submission, i would request you to kindly update the code and share it again.
Thank You 
.
@prashant_ml
Actually when i am submitting my predictions for test data the accuracy is 35% because the labels are mismatched.
And from mismatching I mean that there is a fixed notation that should be used as for labels like “Aerodactyl” should be given label 0 but due to generators it is assigned label 6. That is why model accuracy is low. Anything that can be done like providing custom labels to the generator?
hey buddy , your model isn’t learning anything at all.
If you check the metrics , then you will find same validation metrics.
This shows that your model is not learning anything.
and why have you set all layers as non trainable , kindly change that once.
Only set mobilenet layers to non-trainable not the forward network layers.