EXC_BAD_ACCESS crash when running app with YOLOv3-SPP model trained on custom data.

Ma-Dan / YOLOv3-CoreML

YOLOv3 for iOS implemented using CoreML.

MIT License

171 stars 46 forks source link

EXC_BAD_ACCESS crash when running app with YOLOv3-SPP model trained on custom data. #17

Open fellowProgrammer opened 4 years ago

fellowProgrammer commented 4 years ago

I trained a YOLOv3-SPP model using custom data, I have 3 classes. I then converted it to a Keras .h5 file and then converted it to a CoreML. I ended up with a CoreML model whose Prediction section looks like this.

I tried to use the app by adding the model to Xcode and changing the numClasses variable to 3, yet I still get this crash.

I am quite new to the CoreML world so I don't have a good grasp on what info the MLMultiArrays hold and how to access it etc. What can I do to solve this crash? Is there something wrong with my model or is it just the app?

andriikrupka commented 4 years ago

Any updates? Have the same issue

tj16kimura commented 4 years ago

Me too. Are there any solutions????

tj16kimura commented 4 years ago

I solved this problem by replacing

    for i in 0..<3 {  
            featurePointer = UnsafeMutablePointer<Double>(OpaquePointer(features[i].dataPointer))
            channelStride = features[i].strides[0].intValue
            yStride = features[i].strides[1].intValue
            xStride = features[i].strides[2].intValue

    for i in 0..<3 {  
            featurePointer = UnsafeMutablePointer<Double>(OpaquePointer(features[i].dataPointer))
            channelStride = features[i].strides[2].intValue
            yStride = features[i].strides[3].intValue
            xStride = features[i].strides[4].intValue

thanks

andriikrupka commented 4 years ago

@tj16kimura How did you find that? It's not so obvious

tj16kimura commented 4 years ago

@tj16kimura How did you find that? It's not so obvious

I checked the shape of features and found it was not correct. I think it's because of Xcode's version.

sorry for my cheap English :<