Update app.py

app.py

@@ -7,26 +7,38 @@ import gradio as gr
 SIDE_IMG = 320
 BLACK_TOLERANCE = 20
 
-LIMIT_FOR_DISEASE = 0.8
-
-model = tensorflow.keras.models.load_model('model_1651184375_mejor.h5')
-
-def remove_borders_color(image):
-  image_gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
-  image_gray[image_gray < BLACK_TOLERANCE] = 0
-  y_nonzero, x_nonzero = np.nonzero(image_gray)
-  return image[np.min(y_nonzero):np.max(y_nonzero), np.min(x_nonzero): np.max(x_nonzero), ]
-
-def resize_image_square(image, side):
-  return cv2.resize(image, (side, side), interpolation = cv2.INTER_LINEAR)
+LIMIT_FOR_DISEASE = 0.5
+
+model = tensorflow.keras.models.load_model('predictor_Disease_Risk.h5')
+
+def cut_and_resize(image):
+  LOW_TOL = 20
+  img_bw = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
+  img_bw[img_bw<=LOW_TOL] = 0
+  y_nonzero, x_nonzero = np.nonzero(img_bw)
+  image = image[np.min(y_nonzero):np.max(y_nonzero), np.min(x_nonzero): np.max(x_nonzero), ] / 255.0
+  return cv2.resize(image, SHAPE[:2], interpolation = cv2.INTER_LINEAR)
+
+def z_score(image):
+  img = np.copy(image) / 255.0
+  img_nan = np.copy(img)
+  img_nan[img_nan<0.05] = np.nan
+  _means = np.nanmean(img_nan, axis=(0,1))
+  _stds = np.nanstd(img_nan, axis=(0,1))
+  for i in range(3):
+    if _stds[i] > 0.0:
+      img[:,:,i] = (img[:,:,i] - _means[i]) / _stds[i]
+      img[:,:,i][img[:,:,i]<=img[0,0,i]] = -3.0 
+  img[img>3.0] = 3.0
+  img[img<-3.0] = -3.0
+  return (img + 3.0) / 6.0
 
 def get_image(path):
   image = cv2.imread(path)
-  return resize_image_square(remove_borders_color(image), SIDE_IMG)
-
+  return z_score(cut_and_resize(image))
 
 def predict(path):
-  image = get_image(path) / 255
+  image = get_image(path)
   result = model.predict(np.array([ image ]))[0]
   return 'Retina sana' if result < LIMIT_FOR_DISEASE else 'Retina enferma'