How do you improve the accuracy of point clouds

I am a student currently using the ZED Mini camera. Over the course of three years, I have noticed that with each new SDK release, the accuracy of the point cloud obtained through ROS has been improving, even without any hardware upgrades. In my recent research, I obtained better object detection results by upgrading the SDK (from 3.6 to 3.8). I would like to report this progress to my teacher. Could you please tell me what you did to improve the accuracy of the point cloud? It would be even better if you could provide any related papers or websites for reference.

Hi @kankanzheli
point cloud processing is not performed by the hardware of the camera, but it’s a task of the ZED SDK.
The SDK Team works hardly day-by-day to improve the results of all the modules of the ZED SDK, and improving the quality of the depth map, and consequently, the quality of the point cloud, is always one of the main aspects of our research and development process.

Unfortunately, I cannot provide you with detailed information because what you see is the result of many years of research and development, and the source code and the algorithms used are closed information.

What depth module are you using? In v3.8 we introduced the NEURAL depth mode, are you referring to it?