3

I would like to get the custom objects detected with its frame.

  • I had checked regarding this, I was getting many as ARKit implementations - using that we can get live object detections(ie using the input stream as the camera stream), but my requirement is to get it from a given/static image.

  • I found the latest iOS guidelines/samples using the camera as the input stream which provides the classifications or rectangle or face detection etc.

    • There is my problem is I need to get it from a given image with custom multiple Objects.

      For example, I need to identify the clocks in a given image(Image of a wall with multiple clocks with different types), here I can provide the sample clock images. So I need to get the identified clocks with its frame.

      I had checked using Vision's VNImageBasedRequest with object classification, in which I can get classifications with custom trained CoreML model, but there is not getting the boundingBox property to get the frames, which is available in VNDetectedObjectObservation.

Thanks in advance for your valuable time & response.

Unnikrishnan
  • 158
  • 7
  • Did you find anything using coreML? i heave the same requirement, the objects are getting detected in coreML application however when i use the same model and in project i see issues. Howeever same code works fine on live images but not on still images – Gana Dec 07 '20 at 13:44

1 Answers1

1

You can do it making use of Google Cloud AutoML Vision Object Detection. If you want to detect custom objects you need to create your own custom model dataset and train it. Once the model is trained either you can host it on cloud and make an API call from the iOS app to detect the objects from your static image(online) or you can add the trained model to your project to detect the objects from image(offline). You can find the complete documentation here https://cloud.google.com/vision/automl/object-detection/docs/before-you-begin

If you are able to get it done in any other way please let me know how, even I'm working on similar use case.

Mister Perfect
  • 11
  • 3