The documentation for the "get_projection_transform" function within the "FoVPerspectiveCameras" class, found in the "renderer/cameras.py" file, appears to contain an inconsistency regarding the calculation of s1 and s2. According to the documentation, the variables are defined as follows:
s1 is calculated as (1/\tan(fov/2)),
s2 is calculated as (1/(tan(fov/2) *aspect ratio)).
However, in practice, the definitions for s1 and s2 seem to be reversed:
The documentation for the "get_projection_transform" function within the "FoVPerspectiveCameras" class, found in the "renderer/cameras.py" file, appears to contain an inconsistency regarding the calculation of
s1
ands2
. According to the documentation, the variables are defined as follows:s1
is calculated as (1/\tan(fov/2)),s2
is calculated as (1/(tan(fov/2) *aspect ratio)).However, in practice, the definitions for
s1
ands2
seem to be reversed:for example -
self.compute_projection_matrix(znear, zfar, fov, aspect_ratio=1., degrees=degrees)
tensor([[[109.4238, 0.0000, 0.0000, 0.0000], [ 0.0000, 109.4238, 0.0000, 0.0000], [ 0.0000, 0.0000, 1.0101, -1.0101], [ 0.0000, 0.0000, 1.0000, 0.0000]]])
self.compute_projection_matrix(znear, zfar, fov, aspect_ratio=2., degrees=degrees)
tensor([[[ 54.7119, 0.0000, 0.0000, 0.0000], [ 0.0000, 109.4238, 0.0000, 0.0000], [ 0.0000, 0.0000, 1.0101, -1.0101], [ 0.0000, 0.0000, 1.0000, 0.0000]]])