opencv deep network on android device

asked 2018-08-16 07:03:37 -0500

Brijesh gravatar image

Hi, I am running deep learning network (MobilenetSSDCaffe) on android device using OpenCV deep learning module. But I am getting 2-3 fps only , accuracy is very less and poor performance of object detection. While processing only for 4th frame , I am getting 7-9 fps . Can anybody suggest me ,tunable parameter or optimization of model or any other approach to get best performance . Please find below the details of parameters: IN_WIDTH =300 IN_HEIGHT = 300 IN_SCALE_FACTOR = 0.007843 MEAN_VAL = 127.5 THRESHOLD=0.5

Thanks in advance.

edit retag flag offensive close merge delete

Comments

1

that sounds pretty normal to me. with SSD networks, you can tweak the img size (but not in general with cnn's), but the tradeoff is ofc. less accuracy.

you could also try another network/architecture, like squeezenet.

berak gravatar imageberak ( 2018-08-16 08:08:54 -0500 )edit