Enhancements to process images through PIL and also add RESNET50B

Models/model_host.py

@@ -36,20 +36,34 @@
 
 import keras
 from keras.preprocessing.image import ImageDataGenerator
-from flask import Flask, render_template, request
+from flask import Flask, render_template, send_file, request
 import numpy as np
 import tensorflow
 import sys
 import os
 import io
 import PIL.Image
+import uuid
 
 #init flaskapp
 app = Flask(__name__)
 
-# def save_image(imgData):
+# use this on image passed in to the API
-#     with open('output.jpg','wb') as output:
+def save_image(imgData):
-#         output.write(imgData)
+    filename=str(uuid.uuid4())+'.jpg'
+    print('saving {filename}'.format(filename=filename))
+    with open(filename,'wb') as output:
+        output.write(imgData)
+
+# Use this after processing with PIL
+def save_pil_image(image):
+    filename=str(uuid.uuid4())+'.jpg'
+    print('saving {filename}'.format(filename=filename))
+    imageByteArray = io.BytesIO()
+    image.save(imageByteArray, format='JPEG')
+    imageByteArray = imageByteArray.getvalue()    
+    with open(filename,'wb') as output:
+        output.write(imageByteArray)    
     
 
 def render_index():
@@ -76,21 +90,22 @@ def index():
 def ping():
     return "Alive"
 
-# @app.route('/predict_vgg16', methods=['GET','POST'])
+@app.route('/predict_vgg16', methods=['GET','POST'])
-# def predict_vgg16():
+def predict_vgg16():
-#     test_image=request.get_data()
+    test_image=request.get_data()
-#     test_image = PIL.Image.open(io.BytesIO(test_image))
+    test_image = PIL.Image.open(io.BytesIO(test_image))
+#    save_pil_image(test_image)
 #    test_image = test_image.convert('L')
-#     test_array=keras.preprocessing.image.img_to_array(test_image)
+    test_array=keras.preprocessing.image.img_to_array(test_image)
-#     batch_test_array=np.array([test_array])
+    batch_test_array=np.array([test_array])
-#     predictions=vgg16_model.predict(batch_test_array)
+    predictions=vgg16_model.predict(batch_test_array)
-#     if type(predictions) == list:
+    if type(predictions) == list:
-#         average_prediction = sum(predictions)/len(predictions)
+        average_prediction = sum(predictions)/len(predictions)
-#         threshold_output = np.where(average_prediction > 0.5, 1, 0)
+        threshold_output = np.where(average_prediction > 0.5, 1, 0)
-#     else :
+    else :
-#         threshold_output = np.where(predictions > 0.5, 1, 0)
+        threshold_output = np.where(predictions > 0.5, 1, 0)
-#     response=str(predictions)+'-->'+str(threshold_output)
+    response=str(predictions)+'-->'+str(threshold_output)
-#     return response
+    return response
 
 @app.route('/predict_resnet50', methods=['GET','POST'])
 def predict_resnet50():
@@ -109,6 +124,24 @@ def predict_resnet50():
     response=str(predictions)+'-->'+str(threshold_output)
     return response
 
+# This version expects the image to be of the form (x,x,3).  
+@app.route('/predict_resnet50B', methods=['GET','POST'])
+def predict_resnet50B():
+    print('/predict_resnet50B')
+    test_image=request.get_data()
+    save_image(test_image)
+    test_image = PIL.Image.open(io.BytesIO(test_image))
+    test_array=keras.preprocessing.image.img_to_array(test_image)
+    batch_test_array=np.array([test_array])
+    predictions=resnet50b_model.predict(batch_test_array)
+    if type(predictions) == list:
+        average_prediction = sum(predictions)/len(predictions)
+        threshold_output = np.where(average_prediction > 0.5, 1, 0)
+    else :
+        threshold_output = np.where(predictions > 0.5, 1, 0)
+    response=str(predictions)+'-->'+str(threshold_output)
+    return response
+
 @app.route('/predict_lenet5', methods=['GET','POST'])
 def predict_lenet5():
     print('/predict_lenet5')
@@ -126,13 +159,29 @@ def predict_lenet5():
     response=str(predictions)+'-->'+str(threshold_output)
     return response
 
+# This method is used to process an image through PIL and send it back to the client.  The client can then used this processed image as part of the training data
+# so that the model can adapt to images that are processed through PIL
+@app.route('/process_image', methods=['GET','POST'])
+def process_image():
+    print('/process_image')
+    image=request.get_data()
+    image = PIL.Image.open(io.BytesIO(image))
+    imageByteArray = io.BytesIO()
+    image.save(imageByteArray, format='JPEG')
+    imageByteArray = imageByteArray.getvalue() 
+    print('processed {length} bytes.'.format(length=len(imageByteArray)))     
+    return send_file(io.BytesIO(imageByteArray), mimetype='image/jpeg', as_attachment=True, download_name='%s.jpg' % str(uuid.uuid4()))
+
 
 if __name__ == '__main__':
     resnet50_model_name='../Weights/resnet50.h5'
     resnet50_model = keras.models.load_model(resnet50_model_name)    
 
-    # vgg16_model_name='../Weights/model_vgg16.h5'
+    resnet50b_model_name='../Weights/resnet50B.h5'
-    # vgg16_model=keras.models.load_model(vgg16_model_name) 
+    resnet50b_model = keras.models.load_model(resnet50b_model_name)     
+
+    vgg16_model_name='../Weights/vggnet16.h5'
+    vgg16_model=keras.models.load_model(vgg16_model_name) 
 
     lenet_model_name='../Weights/lenet5.h5'
     lenet_model=keras.models.load_model(lenet_model_name)    

Models/resnet50.py

@@ -38,7 +38,7 @@ from matplotlib import pyplot
 import numpy as np
 import tensorflow
 
-def resnet50(input_shape,classes):
+def resnet50(input_shape,classes,model_name='resnet50'):
     x_input=keras.Input(input_shape)
 
     x=Conv2D(64,(7,7),strides=(2,2),name='conv1')(x_input)
@@ -74,7 +74,7 @@ def resnet50(input_shape,classes):
         x=Dense(classes,activation='sigmoid',name='fc'+str(classes))(x)
     else:
         x=Dense(classes,activation='softmax',name='fc'+str(classes))(x)
-    model=keras.Model(inputs=x_input,outputs=x,name='resnet50')
+    model=keras.Model(inputs=x_input,outputs=x,name=model_name)
     return model