Why the trained model's performance is not the same while running on PC and Android?

asked 2016-04-11 02:36:28 -0500

nikogamulin gravatar image

updated 2016-04-12 02:28:53 -0500

Hi,

I have trained a cascade classifier for a particular object. The classifier is supposed to be used in android app. For testing purposes I've written a python script and the model performance seemed great running it on PC. Then, when I tried to use the same model in android app, the performance was poor, both, in terms of speed and accuracy. Does anyone know what could be the cause for different accuracy of the same model?

Thanks!

edit retag flag offensive close merge delete

Comments

Because of the difference in hardware. When running CMAKE platform specific optimizations are enabled. Also the number of cores of the device will give you an influence. There is simply no comparing between a PC and a mobile device!

StevenPuttemans gravatar imageStevenPuttemans ( 2016-04-11 08:08:29 -0500 )edit

Which kind of performance? Speed or accuracy? If speed, see the above.

Tetragramm gravatar imageTetragramm ( 2016-04-11 21:20:03 -0500 )edit

@Tetragramm thanks for reminding me to clarify the question!

nikogamulin gravatar imagenikogamulin ( 2016-04-12 02:34:08 -0500 )edit
1

Are you using the same camera for both detections (video recorded with the mobile camera process on PC)? Cameras quality can have an impact, also, the codec used for recording/playing could have an impact, as well as the underneath binding for opening the video. For the speed of detection, as said previously, you just can't compare that way...

Mathieu Barnachon gravatar imageMathieu Barnachon ( 2016-04-12 02:57:56 -0500 )edit
1

That was going to be my suggestion as well. Make sure you're both using the same camera, say with recorded video on your phone/tablet, and moving it in similar ways. I had one application that failed rather badly after working fine in testing because when walking with your phone you swing it around in a way I didn't in any of my test video. That blurred everything really badly (because the camera is terrible) and meant it didn't work.

Tetragramm gravatar imageTetragramm ( 2016-04-12 06:54:30 -0500 )edit