We're using a Yolo4 (Pytorch) model to detect different barricades and cordons around some workplaces, the use case being any missing cordons/barricades at any point in time.
We're running our models on RTSP streams with full camera zoom.
However the model is underperforming and not able to detect the missing cordons most of the time.
But while we tested the model on stock videos it worked fine to some extent.
Request if anyone can guide us what can be done in such a situation.
Our options are:
Somehow tune or retrain the current model - In this case request to suggest the parameters to tune both at training and inferencing
Jump to some other models like Yolov5 or Yolov6 or EfficientNet etc. that could detect such barricades and missing barricades from a distance with good amount of accuracy.
Request for your kind and expert suggestions.
Thanks!
