How to run yolo3-trained model on video


#1

Is it possible to run yolo model on video? is there an example to be shared?


#2

Hi @fmr,

Sure, that’s definitely possible. You’d need to loop through the frames of the video and apply the pre-trained YOLO model on each frame. You can then stitch the frames (with bounding boxes added) back into a video at the same frame rate. Other models exist for more temporally consistent results (i.e. less jittering of predictions between frames) such as Flow-Guided Feature Aggregation, but frame by frame predictions are a good place to start.

Check out GluonCV for an example of the YOLO3 model, and you can take inspiration from this script for how to work with video inputs.


#3

Thanks for the link to the script! Yes that’s what I was looking for and I was thinking/ hoping that would officially be a part of GluonCV lib


#4

Well there’s a upcoming GluonCV tutorial on how to apply SSD to webcam video, so this might be useful for your project. Check it out here.