CaDDN in openPCDET webcam question

[kurtwagner19]

Would you provide a guideline on how to test the model with new images from a camera and calibration matrices acquired from opencv?
We are using webcam images and replaced the calibration matrices, however we do not get proper results.
Since we are using single camera, i believe for testing, the model should not require camera extrinsic parameters. (therefore it is confusing, however we assumed any extrinsic parameters should not affect the performance- Though not sure if that is the case).

-A general guideline would be helpful on how to test the pretrained model on a custom RGB camera. What hyper-parameters should be considered to be changed for the system to run properly? what hyper parameters should be tweaked with respect to the camera intrinsic parameters to have better performance?

-Another question would be, is there any work around to test the network on an image without having the intrinsic parameters of the camera? Let's say we have a youtube video, and all we have is the resolution of the image? Do you think the detector is capable of estimating the depth of the objects up to a scale?( meaning the relative positioning of objects in 3D world is estimated up to a scale and then using a reference(such as size of the vehicles) we can rectify the distances.)

Redacted from openPCDET question: open-mmlab/OpenPCDet#844