I am trying to write a program to put a bounding box around dead fish, and not the live ones, in a video. I have minimal data (~5k frames and ~7k objects in total ) and it is VERY low quality (poor resolution, very exposed, lots of partially obscured fish, etc).

I understand there may not be a "Best Object Detection Algorithm" but for my specific problem, can anyone recommend one? Accuracy is key, speed is secondary.

I have tried YOLOv5 with less than adequate results and I am currently looking into using EfficientDet.

  • 1
    $\begingroup$ you should describe more in detail what you tried so far, and provide a couple of examples of images if possible. As you put it now we have to trust you you have bad images (and maybe bad annotations) and that you did everything right when trying yolo. $\endgroup$ Commented Jun 23, 2022 at 13:40
  • $\begingroup$ The YOLOv5 predictions couldn't tell the difference between the dead fish and the time stamp in the corner of the video, so in short, it was unusable. I already made it clear that the images are horrible, and I can tell you with confidence that the annotations are perfect. I do believe I did everything right when using YOLO, I just don't think it was the right algorithm for the job. $\endgroup$
    – user55176
    Commented Jun 23, 2022 at 13:44


You must log in to answer this question.