search

nicholaskajoh / ivy

Video-based object counting software.
MIT License
428 stars 170 forks source link

output #11

Closed prashantkr2601 closed 5 years ago

prashantkr2601 commented 5 years ago

after all the setup then run the code we found video processing too slow. how can improve the video processing or video speed rate

nicholaskajoh commented 5 years ago

The major bottleneck is vehicle detection. Tracking is also computationally expensive when there are many vehicles in a frame at the same time. Running the algorithms on a GPU dramatically improves the performance. I'm still working on optimizations that will make the algorithms run better on less powerful hardware, but ultimately, high-end hardware would yield much better results.

yemikudaisi commented 5 years ago

@nicholaskajoh how about taking advantage of multiple cores ?

nicholaskajoh commented 5 years ago

@yemikudaisi I don't know how multiple cores can be taken advantage of to speed up the system. It would be awesome if there's a way to do that. The vehicle counting process is largely sequential. You have to wait for the next frame to arrive to begin processing and you have to process the current frame fast enough in order to take on the next frame when it arrives. Detection, tracking and counting are sequential as well. You can't track an object you haven't detected or count an object you're not tracking.