I am training a CNN model (for which I tried with both, my custom CNN and VGG19 model) with a dataset with 3 classes. Starting from 1st epoch the loss is staying almost constant. I will attach my colab’s link below for your reference: Colab Link
My training losses for the 1st few epochs looks like this:
Epoch: 0 | Train loss: 1.4142, Train accuracy: 0.3319 | Test loss: 1.1003, Test accuracy: 0.3333
Epoch: 1 | Train loss: 1.1002, Train accuracy: 0.3369 | Test loss: 1.1008, Test accuracy: 0.3333
Epoch: 2 | Train loss: 1.1006, Train accuracy: 0.3344 | Test loss: 1.0991, Test accuracy: 0.3333
Epoch: 3 | Train loss: 1.1006, Train accuracy: 0.3313 | Test loss: 1.0990, Test accuracy: 0.3333
I have tried this same approach for longer epochs but it doesn’t converge. So, for like 50 epochs also the loss and accuracy remains the same.
Even if the loss is reducing, it’s very slow.
I have used the same model on different dataset, and for that also the loss and accuracy remains same for like 100 epochs.