adsdfasdga

main.py

@@ -5,38 +5,48 @@ from src.metrics import IOU
 from src.consts import EPOCHS, STEPS, SEED
 from src.generators import create_generators
 from src.helpers import create_folder
+from src.utils import plot_predictions_grid
 from tensorflow.keras.callbacks import ModelCheckpoint
 import tensorflow as tf
-
+import sys
+import random
+import os
 
 if __name__ == "__main__":
-    model = Unet(num_classes=1).build_model()
-
-    compile_params ={
-        'loss':jaccard_loss(smooth=90), 
-        'optimizer':'rmsprop',
-        'metrics':[IOU]
-        }
-                    
-    model.compile(**compile_params)
-
     model_name = "models/unet.h5"
-    modelcheckpoint = ModelCheckpoint(model_name,
+    if(len(sys.argv) <= 1):
-                                    monitor='val_loss',
+        model = Unet(num_classes=1).build_model()
-                                    mode='auto',
+
-                                    verbose=1,
+        compile_params ={
-                                    save_best_only=True)
+            'loss':jaccard_loss(smooth=90), 
+            'optimizer':'rmsprop',
+            'metrics':[IOU]
+            }
+                        
+        model.compile(**compile_params)
+
+        modelcheckpoint = ModelCheckpoint(model_name,
+                                        monitor='val_loss',
+                                        mode='auto',
+                                        verbose=1,
+                                        save_best_only=True)
 
 
-    train_gen = create_generators('training', SEED)
+        train_gen = create_generators('training', SEED)
-    val_gen = create_generators('validation', SEED)
+        val_gen = create_generators('validation', SEED)
 
-    history = model.fit_generator(train_gen,
+        history = model.fit_generator(train_gen,
-                        validation_data=val_gen,
+                            validation_data=val_gen,
-                        epochs=EPOCHS,
+                            epochs=EPOCHS,
-                        steps_per_epoch=STEPS,
+                            steps_per_epoch=STEPS,
-                        validation_steps = STEPS,
+                            validation_steps = STEPS,
-                        shuffle=True,
+                            shuffle=True,
-    )
+        )
-    create_folder('models', '.')
+        create_folder('models', '.')
-    model.save(filepath=model_name)
+        model.save(filepath=model_name)
+
+    elif(sys.argv[1] == '--predictions'):
+        
+        img_names = [random.choice(os.listdir('./images/rgb/img')) for _ in range(3)]
+        loaded_model = tf.keras.models.load_model("./" + model_name, compile=False)
+        plot_predictions_grid(img_names, loaded_model)

setup.py

@@ -5,13 +5,12 @@ import matplotlib.pyplot as plt
 import time
 import sys
 import random
-from tqdm import tqdm
 
 import cv2 
 
-from consts import SEED, JPG_IMAGES, FC_DIR, FEATURES, MASK_DIR, LABELS, RGB_DIR
+from src.consts import SEED, JPG_IMAGES, FC_DIR, FEATURES, MASK_DIR, LABELS, RGB_DIR
-from helpers import  create_folder, convert_tif_to_jpg, progress_bar
+from src.helpers import  create_folder, convert_tif_to_jpg, progress_bar
-from utils import plot_image_grid
+from src.utils import plot_image_grid
    
 if __name__ == "__main__":
     dirs = os.listdir(FEATURES)

src/consts.py

@@ -1,6 +1,6 @@
 import pandas as pd
-FEATURES ='../data/train_features'
+FEATURES ='./data/train_features'
-LABELS  = '../data/train_labels'
+LABELS  = './data/train_labels'
 JPG_IMAGES = 'images'
 RGB_DIR = "rgb/img"
 FC_DIR = "fc/img"
@@ -11,5 +11,5 @@ BATCH = 8
 IMG_SIZE = (512,512)
 SEED = 7
 
-EPOCHS = 10
+EPOCHS = 5
 STEPS = 10

src/generators.py

@@ -12,7 +12,7 @@ def create_generators(mode='training', seed=1):
   train_datagen = ImageDataGenerator(rescale=1 / 255.0,
                                     horizontal_flip=True,
                                     vertical_flip=True,
-                                  validation_split=0.2)
+                                    validation_split=0.2)
 
   # Provide the same seed and keyword arguments to the fit and flow methods
 

src/utils.py

@@ -1,7 +1,9 @@
 import matplotlib.pyplot as plt
 import cv2
 import os
-from consts import JPG_IMAGES, RGB_DIR, MASK_DIR
+import numpy as np
+from src.consts import JPG_IMAGES, RGB_DIR, MASK_DIR
+import tensorflow as tf
 
 def plot_image_grid(image_names):
 
@@ -23,5 +25,35 @@ def plot_image_grid(image_names):
         plt.imshow(m_img, cmap=None)
 
     #show
+    plt.tight_layout()
+    plt.show()
+
+
+def plot_predictions_grid(image_names, model):
+    #number of img rows
+    n_row= len(image_names)
+
+    plt.subplots(n_row, 3, figsize=(6, 3*n_row))
+    for i, img in enumerate(image_names):
+        r_img = cv2.imread(os.path.join(JPG_IMAGES, RGB_DIR, img))
+        m_img = cv2.imread(os.path.join(JPG_IMAGES, MASK_DIR, img))
+        r_img_ext = np.expand_dims(r_img /255.0, axis=0)
+        pred = model.predict(r_img_ext)
+
+        plt.subplot(n_row, 3, i*3+1)
+        plt.title(f'RGB - {img}')
+        plt.axis('off')
+        plt.imshow(r_img, cmap=None)
+
+        plt.subplot(n_row, 3, i*3+2)
+        plt.title(f'Mask - {img}')
+        plt.axis('off')
+        plt.imshow(m_img, cmap='gray')
+
+        plt.subplot(n_row, 3, i*3+3)
+        plt.title(f'Predicted Mask - {img}')
+        plt.axis('off')
+        plt.imshow(tf.keras.preprocessing.image.array_to_img(pred[0]>0.5), cmap='gray')
+
     plt.tight_layout()
     plt.show()