difference between training and testing

[fsalmasri]

I'm using your model core, training and testing procedure. I only changed the dataloader.

So I'm doing an overfitting of 200 images, using the same images for training and testing.

During training my loss goes to zero. Also, when I inspect the bbx between the prediction and output the are the same.

Only when I do testing,I see a huge difference in values or scales between the values and my validation goes to 0.