Hi, when training the Bonnet model from scratch with the Cityscapes dataset (already preprocessed), it gives me a warning in lines 684, 693 and 693 in abstract_net.py, since it tries to divide by zero. It seems the object pix_hist (assigned in line 1286) contains 0's amongst it's values. This results in the network not being able to train, since it produces a bunch of NaN's. ARe there any known issues with this?
Hi, when training the Bonnet model from scratch with the Cityscapes dataset (already preprocessed), it gives me a warning in lines 684, 693 and 693 in abstract_net.py, since it tries to divide by zero. It seems the object pix_hist (assigned in line 1286) contains 0's amongst it's values. This results in the network not being able to train, since it produces a bunch of NaN's. ARe there any known issues with this?