Misalignment and Distortion Questions

[edexheim]

I would like to revisit why the implementation of rs2_deproject_pixel_to_point is different in rscuda_utils.cuh and rs.cpp as described in #10811. Both versions are used in different places of code, and this inconsistency is very confusing. I have seen significant misalignment issues, and I am wondering if this could be the cause.

Furthermore, I was wondering how there can be large areas of incorrect correspondence such as in the image below.

[MartyG-RealSense]

Hi @edexheim CUDA code will only be used by the RealSense SDK (librealsense) if (a) the computer has an Nvidia graphics GPU chip, like the one on Nvidia Jetson computing boards; and (b) if CUDA support is enabled in the SDK.

If the computer does not have an Nvidia GPU, or if it does but the SDK's CUDA support is not enabled, then the SDK's CUDA routines will not be activated and the non-CUDA routines for processes such as alignment will be used instead.

If CUDA support in the SDK is enabled then CUDA graphics acceleration is automatically applied to YUY2 to RGB color conversion, depth-color alignment and pointcloud generation. No other SDK functions except those three are assisted by the CUDA support.

---

In regard to the problems with the depth data, does the image improve if you are able to use the realsense-viewer tool to reset your camera's calibration to its factory-new defaults using the instructions at #10182 (comment)

[MartyG-RealSense]

hi @edexheim Do you require further assistance with this case, please? Thanks!

[MartyG-RealSense]

Case closed due to no further comments received.