Closed oaishi closed 4 years ago
The idea of having this additional feature was to allow the network to learn something other than all-zeros to fill in regions in the output image that were occluded in the original image.
The inspiration came from Victor lempitsky's work on rendering (https://arxiv.org/abs/1906.08240). However, we didn't use this in the end, as it had limited value.
Hope that helps!
Thank you so much for the clarification and sharing the related paper. It helped me a lot!
Hello, in the z_buffer_layers.py class, there is an optional argument, learn_feature. I was unsure about the functionality of this argument as it is commented out in the forward function. Can you kindly help me to understand this?