Resolution of depth computing

Hi, I am using the ros2 wrapper with setting of:

common_stereo.yaml:
    pub_resolution: ‘CUSTOM’
    pub_downscale_factor: 2

zedx.yaml:
    grab_resolution: 'HD1200'

For this setup, is the depth calculated by the resolution of HD1200 then downscale to 960x600, or should it just be computed on the resolution of downscaled resolution?

When using NEURAL depth modes, the depth map is extracted at a fixed resolution, then upscaled/downscaled to the current grab resolution.
The processing depth resolution depends on the selected model: lower resolution for NEURAL_LIGHT, higher resolution for NEURAL_PLUS.