We could discuss this in private, but I wanted to renew discussion about performance on modest chipsets, since that's all I'm interested in personally.
Previously (can't place Issue) we discussed sampling just the blue component, instead of doing HSL conversion. Maybe that's not a bottleneck on a GPU, but I don't know.
Another thing we've discussed is twin cameras, something about synchronization?
When I awoke today my brain was on about using the command-interface to set the LED light intensities to remove guesswork about which LEDs are which. I reckon even if just two intensities are chosen, it should suffice to identify which are which in the picture.
I thought to make the pattern symmetrical, the lights could alternate. I wonder if alternating the lights would provide enough information to synchronize the cameras.
Beyond this, we have the boxes around the points in the picture. Those provide implicit depth information.
I am thinking if this raw data, center, box, intensity, is sent over UDP then applications have enough information to do the rest themselves I think. That would mean that the machine is not doing any extra work, and it would mean the server is not obligated to do anything that can be prone to error or blame.
We could discuss this in private, but I wanted to renew discussion about performance on modest chipsets, since that's all I'm interested in personally.
Previously (can't place Issue) we discussed sampling just the blue component, instead of doing HSL conversion. Maybe that's not a bottleneck on a GPU, but I don't know.
Another thing we've discussed is twin cameras, something about synchronization?
When I awoke today my brain was on about using the command-interface to set the LED light intensities to remove guesswork about which LEDs are which. I reckon even if just two intensities are chosen, it should suffice to identify which are which in the picture.
I thought to make the pattern symmetrical, the lights could alternate. I wonder if alternating the lights would provide enough information to synchronize the cameras.
Beyond this, we have the boxes around the points in the picture. Those provide implicit depth information.
I am thinking if this raw data, center, box, intensity, is sent over UDP then applications have enough information to do the rest themselves I think. That would mean that the machine is not doing any extra work, and it would mean the server is not obligated to do anything that can be prone to error or blame.