boaerosuke

#import <CoreML/CoreML.h>
#import <Vision/Vision.h>
#import "keras_mnist_cnn.h"

boaerosuke

 //scaled image to 28x28
    UIImage *scaledCanvasImage = [self imageWithImage:self.drawingCanvas.image scaledToSize:CGSizeMake(28, 28)];
    self.imageToDetect = [[CIImage alloc]initWithImage:scaledCanvasImage];
    
    MLModel *ml_model = [[[keras_mnist_cnn alloc] init] model];
    VNCoreMLModel *vnc_core_ml_model = [VNCoreMLModel modelForMLModel: ml_model error:nil];

boaerosuke

    NSDictionary *options_dict = [[NSDictionary alloc] init];
    NSArray *request_array = @[request];
    
    VNImageRequestHandler *handler = [[VNImageRequestHandler alloc] initWithCIImage:self.imageToDetect options:options_dict];
    dispatch_queue_t myCustomQueue;
    myCustomQueue = dispatch_queue_create("com.aname.VNImageRequestHandlerQueue", NULL);
    
    self.resultLabel.text = @"Predicting...";
    dispatch_sync(myCustomQueue, ^{
        [handler performRequests:request_array error:nil];

boaerosuke

    MLModel *ml_model = [[[keras_mnist_cnn alloc] init] model];
    VNCoreMLModel *vnc_core_ml_model = [VNCoreMLModel modelForMLModel: ml_model error:nil];

boaerosuke

 VNCoreMLRequest *request = [[VNCoreMLRequest alloc] initWithModel: vnc_core_ml_model completionHandler: (VNRequestCompletionHandler) ^(VNRequest *request, NSError *error){
        NSArray *results = [request.results copy];
        
        VNCoreMLFeatureValueObservation *res = ((VNCoreMLFeatureValueObservation *)(results[0]));
        
        NSNumber *prediction = [NSNumber numberWithFloat:0];
        NSNumber *compare= [NSNumber numberWithFloat:0];
        int atIndex = 0;
        
        

boaerosuke

//helper method to scale uiimage
- (UIImage *)imageWithImage:(UIImage *)image scaledToSize:(CGSize)newSize {
    UIGraphicsBeginImageContextWithOptions(newSize, NO, 1.0);
    [image drawInRect:CGRectMake(0, 0, newSize.width, newSize.height)];
    UIImage *newImage = UIGraphicsGetImageFromCurrentImageContext();
    UIGraphicsEndImageContext();
    return newImage;
}

boaerosuke

import coremltools
import numpy
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
from keras.utils import np_utils
from keras.models import load_model

boaerosuke

model.save('my_mnist_keras_cnn_model.h5')

boaerosuke

#if K.image_data_format() == 'channels_first':
#    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
#    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
#    input_shape = (1, img_rows, img_cols)

#changed this to be always channels_last!
x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)

boaerosuke

-(void)predictDigit{
  
    //unscaled image
    //self.imageToDetect = [[CIImage alloc]initWithImage:self.drawingCanvas.image];
    
    //scaled image to 28x28
    UIImage *scaledCanvasImage = [self imageWithImage:self.drawingCanvas.image scaledToSize:CGSizeMake(28, 28)];
    self.imageToDetect = [[CIImage alloc]initWithImage:scaledCanvasImage];
    
    MLModel *ml_model = [[[keras_mnist_cnn alloc] init] model];