Training ResNet for 2 class object detection

Hi there,

I’m working on a project and I would like to train a CNN like ResNet-18 for object detection.
I want it to detect two classes: apples and bananas. I’ve processed my dataset into kitti format (5,734 images for training and 1,226 images for validation).

So I uploaded the dataset into DIGITS and started the training (Solver type: Adam, Learning rate: 0.0001 and googlenet pretrained weights)

I trained it for 100 epochs (my god that was long) but at the end… I see 0 for validation on both graphs…

What’s wrong? Maybe 100 epochs isn’t enough?

Thanks for your help!