Stuart Alldritt

@DavidEGrayson @jstroh I agree with both your conclusions! I believe this chunk of code: ``` // Check if a previous frame is using this image (i.e. there is its fence...

Confirmed this is occurring on my ROS Melodic, Ubuntu 18.04 machine. I suspect the main image processing loop is running too slowly on the CPU, since the K4A Viewer does...

Confirmed, the image processing loop take between 40-70ms when the RGB point cloud is enabled. I'll look for ways to potentially speed it up.

The fast point-cloud example does not produce RGB point clouds, just regular point clouds. It's just a way to produce point clouds quickly without using SSE3. My documentation in usage.md...

Thanks for digging into this issue RoseFlunder. I suspect the iterator used to fill up the PointCloud2 message is very expensive. There is almost certainly a faster way to copy...

Going to start having a look at this.

EDIT: **Never mind, the original code has a simple mistake: not calling `pcl_modifier.resize()` before trying to modify the point cloud! Using the sped-up branch helps, but still doesn't get us...

It's good to see that this is using `memset_avx2_erms`: that's probably the compiler optimizing the PointCloud2 alloc.

That looks like a linear interpolation bug. `image_proc/rectify` seems to be using `interpolation=1` (linear interpolation) when it should be using `interpolation=0` (nearest neighbor). Try changing this chunk in the ROS...

FYI: I no longer work at Microsoft, and I wrote this well over a year ago, so my knowledge is a bit fuzzy. As best as I can remember, Azure...