Receives image file as input for prediction

django_project/akangatu/core/predictor.py

@@ -2,6 +2,8 @@
 
 from .preprocessor import Preprocessor
 
+preprocessor = Preprocessor()
+
 class Predictor:
 	_model = None
 
@@ -15,7 +17,7 @@ def predict(self, pixel_array):
 		if not self._model:
 			self._load_model()
 
-		pixels = Preprocessor.reshape_data(pixel_array)
+		pixels = preprocessor.image_to_mnist(pixel_array)
 
 		probs = self._model.predict(pixels)
 		return probs.argmax(axis=1)

django_project/akangatu/core/preprocessor.py

@@ -1,6 +1,39 @@
 import numpy as np
 
+
+MNIST_SIZE = 28
+COLOR_SCALE = 255
+
 class Preprocessor:
-	def reshape_data(request):
-		pixel_array = np.array(request)
-		return pixel_array.reshape(-1, 1, 28, 28) / 255.0
+	def image_to_mnist(self, image):
+		pixel_array = self._aggregate_to_mnist_array(image)
+
+		return self._reshape_data(pixel_array) / COLOR_SCALE
+
+	def _reshape_data(self, pixel_array):
+		return pixel_array.reshape(1, 1, MNIST_SIZE, MNIST_SIZE)
+
+	def _get_group_sizes(self, image):
+		image_width = len(image[0])
+		image_height = len(image)
+
+		return image_height // MNIST_SIZE, image_width // MNIST_SIZE
+
+	def _aggregate_to_mnist_array(self, image):
+		mnist_array = np.zeros([MNIST_SIZE, MNIST_SIZE])
+
+		for i in range(MNIST_SIZE):
+			for j in range(MNIST_SIZE):
+				mnist_array[i][j] = self._calculate_mnist_pixel(image,i,j)
+
+		return mnist_array
+
+	def _calculate_mnist_pixel(self, image,line,column):
+		group_height, group_width = self._get_group_sizes(image)
+
+		rgb_sum = 0
+		for i in range(group_height):
+			for j in range(group_width):
+				rgb_sum += sum(image[line*group_height+i][column*group_width+j])//len(image[line][column])
+
+		return rgb_sum / (group_width * group_height)

django_project/akangatu/views.py

@@ -2,6 +2,7 @@
 from django_quicky import routing
 from django.views.decorators.csrf import csrf_exempt
 import json
+import matplotlib.image as img
 
 from .models import Predictor
 
@@ -25,11 +26,11 @@ def predict(request):
     if request.method.upper() != 'POST':
         return HttpResponseNotAllowed(['POST'])  # List of allowed ones
 
-    json_data = request.POST.dict() or json.loads(request.body.decode('utf-8'))
+    image_file = next(iter(request.FILES.values()))
 
-    print(json_data)
+    image = img.imread(image_file)
 
-    pred = predictor_acessor.predict(json_data['pixels'])
+    pred = predictor_acessor.predict(image)
 
     answer = {'label': int(pred[0]) }
 

requirements.txt

@@ -9,7 +9,9 @@ Keras==2.2.4
 Keras-Applications==1.0.6
 Keras-Preprocessing==1.0.5
 Markdown==3.0.1
+matplotlib==3.0.2
 numpy==1.15.4
+Pillow==5.3.0
 protobuf==3.6.1
 pytz==2018.7
 PyYAML==3.13