Point cloud resolution options

Software
1

Hi,
I wanted to understand the difference between different point cloud resolutions specified in sl_types.hpp

typedef enum
{
  PUB,     //!< Same resolution as Color and Depth Map. [Old behavior for compatibility]
  FULL,    //!< Full resolution. Not recommended because slow processing and high bandwidth requirements
  COMPACT,  //!< Standard resolution. Optimizes processing and bandwidth
  REDUCED   //!< Half resolution. Low processing and bandwidth requirements
} PcRes;
std::string toString(const PcRes & res)
{
  switch (res) {
    case PUB:
      return "PUB";
    case FULL:
      return "FULL";
    case COMPACT:
      return "COMPACT";
    case REDUCED:
      return "REDUCED";
    default:
      return "";
  }
}

const int NEURAL_W = 896;
const int NEURAL_H = 512;

Why is the NEURAL_W and NEURAL_H set to 896 and 512. what do they represent? What was the intention behind having the consts NEURAL_W and NEURAL_H?

Is that the resolution that the depth is originally computed in and then upsampled or downsampled as needed to satisfy the published pointcloud resolution?

2

It is an optimized resolution to improve the computing performance of the AI depth processing pipeline.

This is not completely correct. Unfortunately, I cannot disclose too much information because this is part of the internal processing of the ZED SDK pipeline.
Consider 896x512 a magic setting that enables the processing of depth information to speed up.