I've tried both xception and resnet-101 backbone. The resent-101 backbone is working fine but when using xception backbone, results are very bad. And even, training for longer making the resulting image completely black(after the multiplication of the model output image by value 255). Does anyone know the fix?
I've tried both xception and resnet-101 backbone. The resent-101 backbone is working fine but when using xception backbone, results are very bad. And even, training for longer making the resulting image completely black(after the multiplication of the model output image by value 255). Does anyone know the fix?