Is a D1 sub-stream less accurate than the main stream for AI analysis?

[105437]

Reason I ask is that I recently changed mine from main to sub and I'm noticing that I'm getting a higher percentage of "Nothing Found" from a camera that views the street and captures vehicles using the same version of CPAI and the medium model size. Thanks

 

[wittaj]

Remember the mainstream is downrezed to I believe 640x416, whereas D1 is 720x480.

What is your bitrate for the substream?

I have no issues using D1 substream, even for plates.

 

[105437]

Remember the mainstream is downrezed to I believe 640x416, whereas D1 is 720x480.

What is your bitrate for the substream?

I have no issues using D1 substream, even for plates.

 

[wittaj]

That should be good unless you are stretching the field of view. You could try a higher bitrate just for kicks.

 

[105437]

That should be good unless you are stretching the field of view. You could try a higher bitrate just for kicks.

Would the speed of the vehicle be a factor? Would a higher bit rate help in that scenario?

 

[wittaj]

Depends on the other settings like shutter and what not.

 

[105437]

Depends on the other settings like shutter and what not.

 

[wittaj]

I would drop NR down lower - maybe 40.

What is the size of the vehicle in the field of view?

 

[105437]

I would drop NR down lower - maybe 40.

What is the size of the vehicle in the field of view?

Both NRs to 40? Vehicle crosses the FOV and occupies around 25% of the view.

 

[wittaj]

Yeah that should be good