DeepStack processing times are equal for main stream and sub stream?

[ivanerr]

I've run a few tests and found that processing times are the same for the main stream (1920x1080) and the sub-stream (856x480). I'm attaching the images - two for mode=low and two for mode=high.

I also tried sending a sample image gradually reducing its resolution to the API endpoint and got the same result. Processing times are the same for 1920x1024 and let's say 100x100 pixels.

Am I correct in my assumption that DeepStack resizes the images according to the mode setting and it doesn't really matter what their original sizes are? Or am I missing something? I'm quite puzzled because I heard that people get different results when they switch detection to sub-stream.

Does anyone know what resolution the modes correspond to?

I'm running a Windows GPU version on T400.

 

[MikeLud1]

I've run a few tests and found that processing times are the same for the main stream (1920x1080) and the sub-stream (856x480). I'm attaching the images - two for mode=low and two for mode=high.

I also tried sending a sample image gradually reducing its resolution to the API endpoint and got the same result. Processing times are the same for 1920x1024 and let's say 100x100 pixels.

Am I correct in my assumption that DeepStack resizes the images according to the mode setting and it doesn't really matter what their original sizes are? Or am I missing something? I'm quite puzzled because I heard that people get different results when they switch detection to sub-stream.

Does anyone know what resolution the modes correspond to?

I'm running a Windows GPU version on T400.

If you set the mode to High DS will downsize the image that BI sends it to 640 x 640 before it tries to detect what object is in it. Medium downsize = 416 x 416 and Low downsize = 256 x 256. Below is a part of the code from DeepStack.

DETECTION_HIGH=640,
DETECTION_MEDIUM=416,
DETECTION_LOW=256,
FACE_HIGH=416,
FACE_MEDIUM=320,
FACE_LOW=256,