Model Inference on Video

[yuanyuanli85]

Model Inference on Video

[yuanyuanli85]

To support video, I think the change is going to be something like below. The main thread reads data from video capture, and queue the cvmat. Then start the worker to read from the queue and do processing.

    def start(self, xfilelst):
        # put all of files into queue
        video = cv2.videoCapture()
        while(cap.isOpened()):
            ret, frame = cap.read()
            self._queue.put(frame)
    
        #add a None into queue to indicate the end of task
        self._queue.put(None)

        #start the workers
        for worker in self._workers:
            worker.start()

        # wait all fo workers finish
        for worker in self._workers:
            worker.join()
        print("all of workers have been done")

[rachmadionl]

Thank you so much for your reply!

I am doing a semantic segmentation task, and would like to show the video result of an overlay between original and segmented video. So, in order to do that, should I define it in my worker script or this scheduler? I am rather new with multiprocessing so I don't really get the hang of it. Much appreciated for your kindness!

[yuanyuanli85]

I think you should modify the worker code for your segementation purpose.

[staebchen0]

Hi,
what is needed to be change if I want to read a webcam in real time and call model.predict for the frames?

For the cams I already use a class for the call.
Would it be enough if I just called the predict function?
or what should I change?

Thanks in advance