Initial commit

Tensorflow/concours_drive_2/README.md

@@ -0,0 +1,5 @@
+# Concours DRIVE: Digital Retinal Images for Vessel Extraction
+Mesure de similarité: Dice et Jaccard à la rescousse !
+
+La vidéo du tutoriel se trouve à l'adresse suivante:
+https://www.youtube.com/watch?v=26mpUDOS_IE

Tensorflow/concours_drive_2/model.py

@@ -0,0 +1,87 @@
+import tensorflow as tf
+from tensorflow.keras import layers, models
+
+def LossDice(y_true, y_pred):
+  numerateur  =tf.reduce_sum(y_true*y_pred, axis=(1, 2))
+  denominateur=tf.reduce_sum(y_true+y_pred, axis=(1, 2))
+  dice=2*numerateur/(denominateur+1E-4)
+  return 1-dice
+  
+def LossJaccard(y_true, y_pred):
+  intersection=tf.reduce_sum(y_true*y_pred, axis=(1, 2))
+  union       =tf.reduce_sum(y_true+y_pred, axis=(1, 2))
+  jaccard=intersection/(union-intersection+1E-4)
+  return 1-jaccard
+
+def model(nbr):
+    entree=layers.Input(shape=(576, 560, 3), dtype='float32')
+
+    result=layers.Conv2D(nbr, 3, activation='relu', padding='same')(entree)
+    result=layers.BatchNormalization()(result)
+    result=layers.Conv2D(nbr, 3, activation='relu', padding='same')(result)
+    result1=layers.BatchNormalization()(result)
+
+    result=layers.MaxPool2D()(result1)
+
+    result=layers.Conv2D(2*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+    result=layers.Conv2D(2*nbr, 3, activation='relu', padding='same')(result)
+    result2=layers.BatchNormalization()(result)
+
+    result=layers.MaxPool2D()(result2)
+
+    result=layers.Conv2D(4*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+    result=layers.Conv2D(4*nbr, 3, activation='relu', padding='same')(result)
+    result3=layers.BatchNormalization()(result)
+
+    result=layers.MaxPool2D()(result3)
+
+    result=layers.Conv2D(4*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+    result=layers.Conv2D(4*nbr, 3, activation='relu', padding='same')(result)
+    result4=layers.BatchNormalization()(result)
+
+    result=layers.MaxPool2D()(result4)
+
+    result=layers.Conv2D(8*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+    result=layers.Conv2D(4*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+
+    result=layers.UpSampling2D()(result)
+    result=tf.concat([result, result4], axis=3)
+
+    result=layers.Conv2D(8*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+    result=layers.Conv2D(4*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+
+    result=layers.UpSampling2D()(result)
+    result=tf.concat([result, result3], axis=3)
+    
+    result=layers.Conv2D(4*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+    result=layers.Conv2D(2*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+
+    result=layers.UpSampling2D()(result)
+    result=tf.concat([result, result2], axis=3)
+    
+    result=layers.Conv2D(2*nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+    result=layers.Conv2D(nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+    
+    result=layers.UpSampling2D()(result)
+    result=tf.concat([result, result1], axis=3)
+    
+    result=layers.Conv2D(nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+    result=layers.Conv2D(nbr, 3, activation='relu', padding='same')(result)
+    result=layers.BatchNormalization()(result)
+
+    sortie=layers.Conv2D(1, 1, activation='sigmoid', padding='same')(result)
+
+    model=models.Model(inputs=entree, outputs=sortie)
+    return model

Tensorflow/concours_drive_2/train.py

@@ -0,0 +1,98 @@
+import tensorflow as tf
+from sklearn.model_selection import train_test_split
+from PIL import Image
+import os
+import numpy as np
+import random
+import cv2
+import model
+import traitement_images as ti
+
+dir_images='./training/images/'
+dir_mask  ='./training/1st_manual/'
+
+if not os.path.isdir(dir_images):
+    quit("The directory {} doesn't exist !".format(dir_images))
+if not os.path.isdir(dir_mask):
+    quit("The directory {} doesn't exist !".format(dir_mask))
+
+tab_images=[]
+tab_masks=[]
+
+list_file=os.listdir(dir_images)
+if list_file is None:
+    quit("No file in {} !".format(dir_images))
+    
+for fichier in list_file:
+    img_orig=cv2.imread(dir_images+fichier)
+    tab_images.append(img_orig[:576, :560])
+    num=fichier.split('_')[0]
+    file_mask=dir_mask+num+'_manual1.gif'
+    if not os.path.isfile(file_mask):
+        quit("Mask of {} doesn't exist in {}".format(file_mask, dir_mask))
+    img_mask_orig=np.array(Image.open(file_mask))
+    tab_masks.append(img_mask_orig[:576, :560])
+
+    for angle in range(0, 360, 30):
+        img_r=ti.rotateImage(img_orig, angle)
+        img=img_r.copy()
+        img=ti.random_change(img)
+        tab_images.append(img[:576, :560])
+        img_mask=ti.rotateImage(img_mask_orig, angle)
+        tab_masks.append(img_mask[:576, :560])
+        
+        img=cv2.flip(img_r, 0)
+        img=ti.random_change(img)
+        tab_images.append(img[:576, :560])
+        img_m=cv2.flip(img_mask, 0)
+        tab_masks.append(img_m[:576, :560])
+
+        img=cv2.flip(img_r, 1)
+        img=ti.random_change(img)
+        tab_images.append(img[:576, :560])
+        img_m=cv2.flip(img_mask, 1)
+        tab_masks.append(img_m[:576, :560])
+
+        img=cv2.flip(img_r, -1)
+        img=ti.random_change(img)
+        tab_images.append(img[:576, :560])
+        img_m=cv2.flip(img_mask, -1)
+        tab_masks.append(img_m[:576, :560])
+
+tab_images=np.array(tab_images, dtype=np.float32)/255
+tab_masks =np.array(tab_masks,  dtype=np.float32)[:, :, :]/255
+
+train_images, test_images, train_masks, test_masks=train_test_split(tab_images, tab_masks, test_size=0.05)
+
+del tab_images
+del tab_masks
+
+my_model=model.model(64)
+
+my_model.compile(optimizer='adam',
+                 loss=model.LossDice,
+                 metrics=['accuracy'])
+my_model.fit(train_images,
+             train_masks,
+             epochs=20,
+             batch_size=4,
+             validation_data=(test_images, test_masks))
+
+dir_test_images='./test/images/'
+
+tab_test_images=[]
+tab_files=[]
+for fichier in os.listdir(dir_test_images):
+    img=cv2.imread(dir_test_images+fichier)
+    tab_test_images.append(img[:576, :560])
+    tab_files.append(fichier.split('_')[0])
+
+tab_test_images=np.array(tab_test_images, dtype=np.float32)/255
+tab_files=np.array(tab_files)
+
+for id in range(len(tab_test_images)):
+    mask=np.zeros((584, 565, 1), dtype=np.float32)
+    prediction=my_model.predict(np.array([tab_test_images[id]]))
+    mask[:576, :560]=prediction[0]*255
+    cv2.imwrite("./predictions/"+str(tab_files[id])+".png", mask)
+

Tensorflow/concours_drive_2/traitement_images.py

@@ -0,0 +1,30 @@
+import cv2
+import numpy as np
+import random
+
+def rotateImage(image, angle):
+    image_center=tuple(np.array(image.shape[1::-1])/2)
+    rot_mat=cv2.getRotationMatrix2D(image_center, angle, 1.0)
+    result=cv2.warpAffine(image, rot_mat, image.shape[1::-1], flags=cv2.INTER_LINEAR)
+    return result
+
+def bruit(image):
+    h, w, c=image.shape
+    n=np.random.randn(h, w, c)*random.randint(5, 30)
+    return np.clip(image+n, 0, 255).astype(np.uint8)
+
+def change_gamma(image, alpha=1.0, beta=0.0):
+    return np.clip(alpha*image+beta, 0, 255).astype(np.uint8)
+
+def color(image, alpha=20):
+    n=[random.randint(-alpha, alpha), random.randint(-alpha, alpha),random.randint(-alpha, alpha)]
+    return np.clip(image+n, 0, 255).astype(np.uint8)
+
+def random_change(image):
+    if np.random.randint(2):
+        img=change_gamma(image, random.uniform(0.8, 1.2), np.random.randint(100)-50)
+    if np.random.randint(2):
+        img=bruit(image)
+    if np.random.randint(2):
+        img=color(image)
+    return image