Mask_rcnn: Can this model run at video file or webcam?

[techjjun]

Is there a code for input video image or webcam? and how fast would it be? In paper they said 5fps but is it fast enough to be called real-time?

[isVoid]

My fork uses opencv to feed live images from webcam for prediction. With single Titan Xp it achieves 0.23s/frame (4~5fps). Checkout demo_camera.ipnb

Also see demo vid: https://vimeo.com/258721300

[techjjun]

Thanks for the answer!

[PratibhaT]

@isVoid Have you changed anything on this code? As I used balloon.py as base code and trained it for 3 classes for my own dataset. And for me evaluation is taking like 4sec for 1 image? I'm using TitanX gpu. Can you suggest how can I run it for 5 fps. Also, do you think this network can be made to run at a higher fps?

[Adithia99]

@techjjun do you the codes? how to detect with fps? i need that please share it