How to use nvdec/nvenc

[johnnynunez]

Issue guidelines

• I've read the Issue Guidelines and wholeheartedly agree.

Issue Checklist

• I have searched open or closed issues for my problem and found nothing related or helpful.
• I have read the Documentation and found nothing related to my problem.
• I have gone through the Bonus Examples and FAQs and found nothing related or helpful.

Describe your Question

I am trying to make a yolo detector using mp4 video or real time camera. I am trying to speed up my pipeline because the fps of the model is much higher than the video and the fps marked. How can I use hardware e.g. nvenc and nvenc to encode and decode the video?
I have opencv and ffmpeg with cuda. The device is a jetson agx orin

Terminal log output(Optional)

No response

Python Code(Optional)

No response

VidGear Version

0.3.2

Python version

3.11

Operating System version

Ubuntu 22.04

Any other Relevant Information?

No response

[welcome]

Thanks for opening this issue, a maintainer will get back to you shortly!

In the meantime:

• Read our Issue Guidelines, and update your issue accordingly. Please note that your issue will be fixed much faster if you spend about half an hour preparing it, including the exact reproduction steps and a demo.
• Go comprehensively through our dedicated FAQ & Troubleshooting section.
• For any quick questions and typos, please refrain from opening an issue, as you can reach us on Gitter community channel.

[abhiTronix]

I am trying to make a yolo detector using mp4 video or real time camera. I am trying to speed up my pipeline because the fps of the model is much higher than the video and the fps marked. How can I use hardware e.g. nvdec and nvenc to encode and decode the video?
I have opencv and ffmpeg with cuda. The device is a jetson agx orin

@johnnynunez Don't use OpenCV for video IO, but FFmpeg directly which supports both nvenc and nvenc to encode and decode the video. For this purpose, You could use my deffcode library instead which will be integrated soon in #148. The complete HW transcoding example is given in this docs: https://abhitronix.github.io/deffcode/latest/recipes/advanced/transcode-hw-acceleration/#hardware-accelerated-video-transcoding

Goodluck!

[johnnynunez]

I am trying to make a yolo detector using mp4 video or real time camera. I am trying to speed up my pipeline because the fps of the model is much higher than the video and the fps marked. How can I use hardware e.g. nvenc and nvenc to encode and decode the video?
I have opencv and ffmpeg with cuda. The device is a jetson agx orin

@johnnynunez Don't use OpenCV for video IO, but FFmpeg directly which supports both nvenc and nvenc to encode and decode the video. For this purpose, You could use my deffcode library instead which will be integrated soon in #148. The complete HW transcoding example is given in this docs: https://abhitronix.github.io/deffcode/latest/recipes/advanced/transcode-hw-acceleration/#hardware-accelerated-video-transcoding

Goodluck!

but Does it have threaded-queue-mode too?? https://abhitronix.github.io/vidgear/v0.3.2-stable/bonus/TQM/#threaded-queue-mode

class FrameCapture:
    def __init__(self, source=0, stabilize=False, stream_mode=False, logging=False):
        self.source = source
        if isinstance(self.source, int):
            # Use CamGear for optimized live stream handling
            self.vcap = CamGear(source=self.source, stream_mode=stream_mode, logging=logging)
            height, width, _ = self.vcap.frame.shape
            self.fps = self.vcap.framerate
        else:
            # Use VideoGear for general video file handling
            self.vcap = VideoGear(source=self.source, stabilize=stabilize, stream_mode=stream_mode, logging=logging)
            height, width, _ = self.vcap.stream.frame.shape
            self.fps = self.vcap.stream.framerate
        self.video_info = sv.VideoInfo.from_video_path(self.source)
        self.frame_count = 0
        self.frame_size = (width, height)
        self.stopped = False

    def start(self):
        self.stopped = False
        self.vcap.start()

    def read(self):
        if not self.stopped:
            frame = self.vcap.read()
            if frame is None:
                self.stop()
                return None
            self.frame_count += 1
            return frame
        return None

    def get_frame_count(self):
        return self.frame_count

    def get_fps(self):
        return self.fps

    def get_frame_size(self):
        return self.frame_size

    def get_stop(self):
        return self.stopped

    def stop(self):
        # Safely close the video stream
        self.frame_count = 0
        # self.vcap.release()
        self.vcap.stop()
        self.stopped = True


if __name__ == '__main__':
    # Example usage:
    stream = FrameCapture(0)  # 0 for default webcam
    stream.start()

    try:
        input("Press Enter to stop...")
    finally:
        stream.stop()
        print("Exiting program")

[abhiTronix]

You don't need multi-threading here.

[johnnynunez]

You don't need multi-threading here.

and deffcode uses queue? Or only decoding?

[abhiTronix]

and deffcode uses queue? Or only decoding?

@johnnynunez No queue, just decoding. You could learn more about it here: https://abhitronix.github.io/deffcode/latest/recipes/advanced/decode-hw-acceleration/#hardware-accelerated-video-decoding

[johnnynunez]

and deffcode uses queue? Or only decoding?

@johnnynunez No queue, just decoding. You could learn more about it here: https://abhitronix.github.io/deffcode/latest/recipes/advanced/decode-hw-acceleration/#hardware-accelerated-video-decoding

Yeah it is working for me