Equation 1 in PackNet paper described 2 masks denoted as M_p and M_t, which are designed to bypass the loss calculation for the pixels which either don't have a valid mapping or are part of the static scenes and dynamic objects with no relative motion. However, when I go through the scripts in packnet_sfm/losses folder, I found no implementation for such binary masks under unsupervised setup. For unsupervised setup, the only codes I can see related to 'mask' are in forward function declared in packnet_sfm/losses/multiview_photometric_loss.py. It seems like the function only implements equation (3) described in PackNet paper, but never gets equation (1) implemented. Is it this an upcoming feature?
Equation 1 in PackNet paper described 2 masks denoted as M_p and M_t, which are designed to bypass the loss calculation for the pixels which either don't have a valid mapping or are part of the static scenes and dynamic objects with no relative motion. However, when I go through the scripts in packnet_sfm/losses folder, I found no implementation for such binary masks under unsupervised setup. For unsupervised setup, the only codes I can see related to 'mask' are in forward function declared in packnet_sfm/losses/multiview_photometric_loss.py. It seems like the function only implements equation (3) described in PackNet paper, but never gets equation (1) implemented. Is it this an upcoming feature?