Object Detection MLModel for iOS with output configuration of confidence scores & coordinates for the bounding box.

[ajurav1]

I have exported the mlmodel from export.py, and the model exported have a type of Neural Network and output configuration is
[<VNCoreMLFeatureValueObservation: 0x281d80240> 4702BA0E-857D-4CE6-88C1-4E47186E751F requestRevision=1 confidence=1.000000 "2308" - "MultiArray : Float32 1 x 3 x 20 x 20 x 85 array" (1.000000), <VNCoreMLFeatureValueObservation: 0x281d802d0> AE0A0580-7DA2-4991-98BB-CD26EE257C7A requestRevision=1 confidence=1.000000 "2327" - "MultiArray : Float32 1 x 3 x 40 x 40 x 85 array" (1.000000), <VNCoreMLFeatureValueObservation: 0x281d80330> 0253FD2B-10B0-4047-A001-624D1864D27C requestRevision=1 confidence=1.000000 "2346" - "MultiArray : Float32 1 x 3 x 80 x 80 x 85 array" (1.000000)]

I was looking for output type VNRecognizedObjectObservation in yoloV5 instead of VNCoreMLFeatureValueObservation.

So, my question is what information does this VNCoreMLFeatureValueObservation MultiArray hold (is it something like a UIImage or CGRect?, or something different?) and how can I convert this Multidimensional Array into a useful set of data that I can actually see confidence scores & coordinates?

[github-actions]

Hello @ajurav1, thank you for your interest in our work! Please visit our Custom Training Tutorial to get started, and see our Jupyter Notebook , Docker Image, and Google Cloud Quickstart Guide for example environments.

If this is a bug report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom model or data training question, please note that Ultralytics does not provide free personal support. As a leader in vision ML and AI, we do offer professional consulting, from simple expert advice up to delivery of fully customized, end-to-end production solutions for our clients, such as:

• Cloud-based AI systems operating on hundreds of HD video streams in realtime.
• Edge AI integrated into custom iOS and Android apps for realtime 30 FPS video inference.
• Custom data training, hyperparameter evolution, and model exportation to any destination.

For more information please visit https://www.ultralytics.com.

[dlawrences]

Hi @ajurav1

Those tensors store network predictions that are not decoded. There are a few recommendations I made in #343 to decode these and there is even a NumPy sample code that does this for the ONNX model.

There's also guidance here: https://docs.ultralytics.com/yolov5/tutorials/model_export

Take a look in there and let me know if you require further support.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[ShreshthSaxena]

Hi @dlawrences ,
I'm in the same boat trying to figure out how to convert these raw predictions to bbox coordinates. Is there a swift implementation available to interpret these outputs or someway I can add the post-processing to a CoreML pipeline and get the final outputs in Swift?

[dlawrences]

@ShreshthSaxena , there is a lot of useful information which you will be able to use on this blog: https://machinethink.net/blog/

I recommend acquiring the book as well.

[maidmehic]

Hi @ajurav1 and @ShreshthSaxena, have you managed to convert VNCoreMLFeatureValueObservation MultiArrays to some useful info on Swift side?

[kir486680]

@dlawrences In the article, the author says that the "Ultralytics YOLOv5 model has a Core ML version but it requires changes before you can use it with Vision." Did you manage to make it work with Vision?

[dlawrences]

@kir486680 the CoreML model, as exported from this repository, presents the final feature maps, thus the predictions are not decoded, nor any NMS is applied.

I have been able to create the required steps locally, yes, and it works directly with Vision.

[kir486680]

@dlawrences could you please share at least some clues pls? I implemented NMS but I do not know what to do with the output from the model. I seen some implementations here . Do you have something similar?

[wmcnally]

@kir486680 @dlawrences any update on this?

[Workshed]

If you come across this issue, there's a script here for creating a CoreML model which outputs the expected values https://github.com/Workshed/yolov5-coreml-export

[glenn-jocher]

@Workshed thanks for sharing your script for creating a CoreML model that outputs the expected values with YOLOv5. This will be helpful for others who are looking to work with YOLOv5 in CoreML. Keep up the great work!