Coordinates from object detection Core ML model - swift

Question

I created an ML Model for simple object detection. When I used it in the Xcode "Preview" tab, it perfectly identified and put a bounding box around the object. However, when I try to do it programmatically, I end up with an MLMultiArray (or a similar type depending on what I try) which I cannot use to create a bounding box. Here is the relevant part of my code:
func performAnalysis(frame: CVImageBuffer) {
guard let input = try? IdentifyBoomInput(imagePath: frame) else { return }
guard let result = try? identifyBoom.prediction(input: input) else { return }
if result.coordinates.count == 0 { return }
var coords = result.***????????????????***
print(coords)
}
I've tried every member of result (see here), but I'm unable to get anything useful for creating a bounding box. Any help would be greatly appreciated.
Update:
I had originally created the model using CreateML. I thought that maybe CreateML was the issue, so I used PyTorch with YOLOv8. When that didn't work, I tried YOLOv5, which also didn't work. My annotations and training data are clearly fine– as is my original model– because the interface Xcode provides for testing allows me to use it just fine. Thoughts?

Answer 1

Did you take a look at the .mlmodel specification in XCode? Screenshot of my .mlmodel details. I have trained the model with YOLOv5 PyTorch and converted the .pt file into .ml file format.
After that, in the image
3 is how I get the element of my MLMultiArray (1 x 25200 x 5+C), in my situation I only have one class --> (1 x 25200 x 6). As I read in this thread https://github.com/ultralytics/yolov5/issues/7011 :
private func getOutput(image: UIImage) {
let image = image.resizeImageTo(size: CGSize(width: 640, height: 640))
let buffer = image!.convertToBuffer()
let output = try? model.prediction(image: buffer!)
let prediction = output!.var_903
x = Double(truncating: prediction[0])
y = Double(truncating: prediction[1])
width = Double(truncating: prediction[2])
height = Double(truncating: prediction[3])
Hopefully, this could help