I computed HFOVs from the rectified camera matrices provided in calib.npy with the following equation:
$HFOV = 2 * \text{atand}(n_x / (2 f_x))$
where:
- atand: inverse tangent in degrees
- n_x: image width
- f_x: horizontal focal length (in units of pixels)
this yields:
- thermal left: 79 degrees (vs 45 in the paper)
- rgb left: 77 degrees (vs 82.2 in the paper)
- nir left: 90 degrees (vs 69 in the paper)
I checked the one camera for which I could confidently find the HFOV, the left NIR camera from the Intel RealSense D435i, and the spec sheet lists the HFOV as 91.2 degrees (see table 3-13). It appears the 69 degree HFOV listed in the paper was taken from either the color sensor (table 3-16) or the standard imager (table 3-12), while the D435i uses the wide imager (as listed in table 3-11).
My guess is that HFOV listed for the RGB camera in the paper is correct and that the rectified RGB images lose a slight bit of that FOV. But it seems the thermal and NIR FOVs listed in the paper are incorrect. Can you confirm whether this is the case?
I computed HFOVs from the rectified camera matrices provided in
$HFOV = 2 * \text{atand}(n_x / (2 f_x))$
calib.npywith the following equation:where:
this yields:
I checked the one camera for which I could confidently find the HFOV, the left NIR camera from the Intel RealSense D435i, and the spec sheet lists the HFOV as 91.2 degrees (see table 3-13). It appears the 69 degree HFOV listed in the paper was taken from either the color sensor (table 3-16) or the standard imager (table 3-12), while the D435i uses the wide imager (as listed in table 3-11).
My guess is that HFOV listed for the RGB camera in the paper is correct and that the rectified RGB images lose a slight bit of that FOV. But it seems the thermal and NIR FOVs listed in the paper are incorrect. Can you confirm whether this is the case?