Why inverse depth divided by 1 ?
AsDeadAsADodo opened this issue · comments
In R-MVSNET training loss section , we apply the inverse depth to sample the depth values in order to efficiently handle reconstructions with wide depth ranges.
In homography_warping.py line 66-69
inv_depth_start = tf.reshape(tf.div(1.0, depth_start), [])
inv_depth_end = tf.reshape(tf.div(1.0, depth_end), [])
inv_depth = tf.lin_space(inv_depth_start, inv_depth_end, depth_num)
depth = tf.div(1.0, inv_depth)
Correct me if I'm wrong , basically did something like this
Why ? is this a way to draw the network's attention to focus on the nearest planes?
Any help would be great !