I want to ask for someone's yolo_net pretrain model, not the yolo_tiny_net

[planckztd]

When I change the yolo_tiny_net into yolo_net, and run the training without pretrain model, it

assert not np.isnan(loss_value), 'Model diverged with loss = NaN' AssertionError: Model diverged with loss = NaN.

I want to know how to deal with this situation.

[planckztd]

OK， I had worked it out. maybe the error caused by losses change too many. Add batch normalization after conv layers or fully connected layers to avoid the loss to be a large number.

[zsano1]

So what is the purpose of pre-train? Can i cut it down?