Showing only one bounding box per frame (coreml)

[1omarsaid]

I am trying to find a way to only show one bounding box per frame, meaning that I dont want it to detect three "persons" in one frame. In the view controller I have modified the maxInflightBuffers to 1 but I am not sure if that is the way to go. The reason of such implementation, I am enabling an audio file when a certain bounding box appears, but if there is more than one, it wont play.

[hollance]

That is unrelated to the maxInflightBuffers. You need to change the nonMaxSuppression code to sort the bounding boxes by confidence and then pick the one with the highest confidence.

[1omarsaid]

How can I do that? so far I just have it so it shows the bounding boxes with a prediction above 50%

[1omarsaid]

@hollance I would really appreciate it if you can give me some guidance into implementing this. Thank you in advance.

[hollance]

There’s not much to it: when you get the bounding boxes, sort them by confidence score and keep the one with the highest score.

[1omarsaid]

What if I just do something like this "return nonMaxSuppression(boxes: predictions, limit: 1, threshold: iouThreshold)" would this be wrong?

[hollance]

That's fine but a little bit slower than my previous suggestion.

[1omarsaid]

How would I be able to reduce the number of bounding boxes (per second) without reducing the "Desired frame rate"

[hollance]

You would have to change the model for that. It always predicts the same number of bounding boxes in every frame.

[1omarsaid]

sorry what I meant was, how can I have a tiny delay between each bounding box, so if I want to associate a bounding box with a half second audio, it doesnt cut off because it shows a new bounding box that also activates the audio file

[1omarsaid]

Do you know if there is a way to do that? I tried setting up delays of when they would trigger the "show" function but it doesnt seem to work.

[hollance]

Every new video frame gives completely new predictions. There is no way to track an object across multiple frames with YOLO. So I’m not really sure what you’re trying to accomplish but it sounds like it needs more than just bounding box detection.

[1omarsaid]

I see, so there isnt a way to be able to only get predictions on every other frame?

[hollance]

If you want predictions every other frame, then simply keep a counter and only do a prediction when the counter is even.

[1omarsaid]

would the counter be in the viewcontroller? Please give me some tips in how I can execute that? what if I want to make it make a prediction every n frames?

[1omarsaid]

which function does the actual prediction? I can wrap it with an if statement this way.

[1omarsaid]

I created an if statement around

      } else {
        boundingBoxes[i].hide()
      }```

it seemed to do the job