Camera Intrinsics

[zichenwang01]

Hi,

I am having a bit of trouble declaring a camera with the intrinsic matrix.

If I understand correctly, the to_world transformation corresponds to the extrinsic matrix, but I will need to compute the fov and other parameters as an equivalent of the intrinsic matrix.

For example, if I have a camera with an image resolution 1600*1200 and an intrinsic matrix

[ [ 2.8923e+03, -5.5663e-04,  8.2321e+02,  0.0000e+00],
  [ 0.0000e+00,  2.8832e+03,  6.1907e+02,  0.0000e+00],
  [ 0.0000e+00,  0.0000e+00,  1.0000e+00,  0.0000e+00],
  [ 0.0000e+00,  0.0000e+00,  0.0000e+00,  1.0000e+00] ]

How should I declare such a camera in Mitsuba? Huge thanks in advance!

Edit: I get that I can compute

fovx = 2 * math.atan(image_width / (2 * fx)) * (180 / math.pi)
fovy = 2 * math.atan(image_height / (2 * fy)) * (180 / math.pi),

but how should I specify fovx, fovy (as opposed to a single fov) and the optical center cx, cy in Mitsuba?

[Start1er]

As I have known about mitsuba3, it use perspective projection matrix convert from 3d space to image plane, which means that unlike intrinsics matrix, it assums that fx=fy so if you specify fov_x, then you dont need to specify fov_y. cx and cy you can find in perspective camera plugin to specify offset_principle_x and offset_principle_y as a percentage rather than pixel as intrinsics

[merlinND]

Hello @zichenwang01,

Please have a look at the documentation of the perspective sensor, which describes the extrinsic (to_world) and intrinsic (fov, fov_axis / focal_length, principal_point_offset_*) parameters.
https://mitsuba.readthedocs.io/en/stable/src/generated/plugins_sensors.html#perspective-pinhole-camera-perspective

[zichenwang01]

I see. Thanks!