TinyYolo Tiny gives long process time once in a while

asked 2019-05-15 06:30:20 -0600

Shay Weissman gravatar image

##### System information (version) - OpenCV => 3.4.5 - Operating System / Platform => Windows 10 64 Bit - Compiler => Visual Studio 2015

Detailed description

. I trained tiny yolo 3 for car detection and use it successfully. In order to reduce process time I shrinked the tiny yolo model. The model recognize well and run faster (~100 ms on my pc) on > 95% of images but from time to time the inference takes much longer up to 5 seconds. It is not the same image that take long process time so it is rather related to the implementation. model at: https://www.dropbox.com/sh/odk7hglm2r...

edit retag flag offensive close merge delete

Comments

It is not the same image that take long process time.

any chance, you're changing the (network) input resolution for new images ?

(it does not like that at all, because the whole network has to be resized)

berak gravatar imageberak ( 2019-05-15 09:23:35 -0600 )edit
2

Hi, I do not change the image size between images- always 416x416. I added the original tiny yolo v3 so you can compare the 2 cfg files, to see what I changed.
https://www.dropbox.com/sh/odk7hglm2r... Thanks.

Shay Weissman gravatar imageShay Weissman ( 2019-05-16 02:32:10 -0600 )edit

Please provide some detailed report about efficiency. At least the inference time measurements.

dkurt gravatar imagedkurt ( 2019-05-22 07:36:58 -0600 )edit