result dir added, learning rate fixed

Author: Sekhar Roy

ReadME_Run_melNET.pdf

@@ -11,3 +11,5 @@
 09) Set �Python Interpreter� to the new Conda Environment (E.g.:melNET_env)
 10) Go to Run >> Run >> melNET  
 11) Follow the prompt! 
+
+N.B.: Follow the provided pdf for more detail

ReadME_Run_melNET.txt

@@ -285,6 +285,8 @@ def file_naming(_img, process_initial, _folder, _set_counter):
     if rotationAngle is not None:
         # Rotation Code
         for angle in range(0, 360, rotationAngle):
+            """
+            # No Artifact Code:
             _scaleFactor = 1
             (h, w) = _img.shape[:2]
             (cX, cY) = (w // 2, h // 2)
@@ -307,6 +309,14 @@ def file_naming(_img, process_initial, _folder, _set_counter):
 
             _imgRotated = cv2.warpAffine(_img, M, (nW, nH), flags=cv2.INTER_LINEAR,
                                          borderMode=cv2.BORDER_REFLECT_101)
+            """
+            # Artifact Code:
+            dim = _img.shape
+            _scaleFactor = 1
+            rotationMatrix = cv2.getRotationMatrix2D((dim[1] / 2, dim[0] / 2), angle, _scaleFactor)
+            _imgRotated = cv2.warpAffine(_img, rotationMatrix, (dim[1], dim[0]), flags=cv2.INTER_LINEAR,
+                                         borderMode=cv2.BORDER_REFLECT_101)
+
             _savingName = dst_root + _folder + "/" + _folder + "_" + str(name_counter) + "_" \
                                    + process_initial + "_Rot_" + str(angle) + ".jpg"
             cv2.imwrite(_savingName, _imgRotated)

__pycache__/blur_detection.cpython-36.pyc

@@ -65,17 +65,18 @@ def quit_button_applied():
         fold_num = 5
         if five_fold:
             for fold in range(fold_num):
-                root = os.getcwd() + "/aug_Data/Five_Fold_(Aug)/Fold_"+str(fold+1)+"/Test"
+                root = os.getcwd() + "/Results/Fold_0"+str(fold+1)
                 result_path = root + "/_result.csv"
                 conf_mat_path = root + "/conf_mat.csv"
                 error_path = root + "/error_analysis.csv"
                 err_thresh = 70
                 conf_mat_make(result_path, conf_mat_path, error_path, err_thresh, pos_class, neg_class)
 
         else:
-            result_path = os.getcwd() + "/aug_Data/Single_Fold_(Aug)/Test/_result.csv"
-            conf_mat_path = os.getcwd() + "/aug_Data/Single_Fold_(Aug)/Test/conf_mat.csv"
-            error_path = os.getcwd() + "/aug_Data/Single_Fold_(Aug)/Test/error_analysis.csv"
+            root = os.getcwd() + "/Results/Fold_01"
+            result_path = root + "/_result.csv"
+            conf_mat_path = root + "/conf_mat.csv"
+            error_path = root + "/error_analysis.csv"
             err_thresh = 70
             conf_mat_make(result_path, conf_mat_path, error_path, err_thresh, pos_class, neg_class)
 

__pycache__/melNET_augData.cpython-36.pyc

@@ -53,12 +53,19 @@ def quit_button_applied():
         accessories_path = os.getcwd() + "/Accessories"   # same name in malNET_train_src
         aug_data_path = os.getcwd() + "/aug_Data"
 
+        result_root = os.getcwd() + "/Results"
+        if tf.gfile.Exists(result_root):
+            tf.gfile.DeleteRecursively(result_root)
+        tf.gfile.MakeDirs(result_root)
+
         # Starting Test
         if tf.gfile.Exists(accessories_path) and tf.gfile.Exists(aug_data_path):
             if five_fold:
                 fold_num = 5  # Five-Fold
                 root = aug_data_path + "/Five_Fold_(Aug)"
                 for num in range(fold_num):
+                    result_path = result_root + "/Fold_0" + str(num+1)
+                    tf.gfile.MakeDirs(result_path)
                     test_path = root + "/Fold_" + str(num+1) + "/Test"
                     weight_path = accessories_path + "/Five_Fold_Trained/Trained_Fold_" + str(num+1)
                     delete_all_csv(test_path)  # Deleting old results
@@ -68,14 +75,16 @@ def quit_button_applied():
                             ground_truth = re.sub(r'[^a-z0-9]+', ' ', str(folder).lower())
                             class_path = test_path + "/" + folder
                             classes = melNET_test_src.main(class_path, weight_path, save_performance, show_detection,
-                                                           ground_truth, test_path)
+                                                           ground_truth, result_path)
                         if save_performance:
-                            merge_csv(test_path)
+                            merge_csv(result_path)
                     else:
                         print("Either [Test Data] or [Weight] is NOT available!")
                         exit()
 
             else:  # Single-Fold
+                result_path = result_root + "/Fold_01"
+                tf.gfile.MakeDirs(result_path)
                 root = aug_data_path + "/Single_Fold_(Aug)"
                 test_path = root + "/Test"
                 weight_path = accessories_path + "/Single_Fold_Trained"
@@ -86,7 +95,7 @@ def quit_button_applied():
                         ground_truth = re.sub(r'[^a-z0-9]+', ' ', str(folder).lower())
                         class_path = test_path + "/" + folder
                         classes = melNET_test_src.main(class_path, weight_path, save_performance, show_detection,
-                                                       ground_truth, test_path)
+                                                       ground_truth, result_path)
                     if save_performance:
                         merge_csv(test_path)
                 else:

__pycache__/melNET_blur_detection.cpython-36.pyc

@@ -28,17 +28,16 @@ def quit_button_applied():
 
     # Variables
     _five_fold = BooleanVar()
-    _learning_rate = IntVar(0)
+    _learning_rate1 = BooleanVar()
+    _learning_rate2 = BooleanVar()
+    _learning_rate3 = BooleanVar()
+    _learning_rate4 = BooleanVar()
     _batch_size = IntVar(0)
     _iteration = IntVar(0)
 
     # UI
     Checkbutton(root, text="Five-Fold (Default: Single-Fold)", variable=_five_fold).grid(row=1, column=3, sticky=W)
 
-    Label(root, text="Learning Rate: ").grid(row=2, column=0)
-    Label(root, text="(Default: 0.01)").grid(row=3, column=0)
-    Entry(root, textvariable=_learning_rate).grid(row=2, column=1)
-
     Label(root, text="Batch Size: ").grid(row=4, column=0)
     Label(root, text="(Default: 100)").grid(row=5, column=0)
     Entry(root, textvariable=_batch_size).grid(row=4, column=1)
@@ -47,32 +46,62 @@ def quit_button_applied():
     Label(root, text="(Default: 4000)").grid(row=7, column=0)
     Entry(root, textvariable=_iteration).grid(row=6, column=1)
 
-    Button(root, text="Quit", command=quit_button_applied, width=15).grid(row=8, column=2, sticky=W)
-    Button(root, text="Apply", command=apply_button_applied, width=15).grid(row=8, column=3, sticky=W)
+    Checkbutton(root, text="Learning Rate: 0.005", variable=_learning_rate1).grid(row=8, column=0, sticky=W)
+    Checkbutton(root, text="Learning Rate: 0.001", variable=_learning_rate2).grid(row=9, column=0, sticky=W)
+    Checkbutton(root, text="Learning Rate: 0.01", variable=_learning_rate3).grid(row=10, column=0, sticky=W)
+    Checkbutton(root, text="Learning Rate: 0.015", variable=_learning_rate4).grid(row=11, column=0, sticky=W)
+    Label(root, text="If NONE or more than one is chosen, default Learning Rate is 0.01").grid(row=12, column=0)
+
+    Button(root, text="Quit", command=quit_button_applied, width=15).grid(row=14, column=2, sticky=W)
+    Button(root, text="Apply", command=apply_button_applied, width=15).grid(row=14, column=3, sticky=W)
 
     root.mainloop()
 
     if applied:
         five_fold = _five_fold.get()
-        learning_rate = _learning_rate.get()
+
+        learning_rate1 = _learning_rate1.get()
+        learning_rate2 = _learning_rate2.get()
+        learning_rate3 = _learning_rate3.get()
+        learning_rate4 = _learning_rate4.get()
+
         batch_size = _batch_size.get()
         iteration = _iteration.get()
 
         # Default Values
-        if learning_rate is 0:
-            learning_rate = 0.01
         if batch_size is 0:
             batch_size = 100
         if iteration is 0:
             iteration = 4000
 
+        if learning_rate1:
+            learning_rate = 0.005
+        elif learning_rate2:
+            learning_rate = 0.001
+        elif learning_rate3:
+            learning_rate = 0.01
+        elif learning_rate4:
+            learning_rate = 0.015
+
+        learning_rate_sum = int(learning_rate1) + int(learning_rate2) + int(learning_rate3) + int(learning_rate4)
+
+        if learning_rate_sum > 1 or learning_rate_sum == 0:
+            learning_rate = 0.01
+
         # Managing directories
         accessories_path = os.getcwd() + "/Accessories"   # same name in malNET_train_src
         if tf.gfile.Exists(accessories_path):
             tf.gfile.DeleteRecursively(accessories_path)
 
         tf.gfile.MakeDirs(accessories_path)
 
+        # Save hyper-parameters
+        hyper_path = accessories_path + "/hyper_parameters.csv"
+        hyper = open(hyper_path, 'w')
+        hyper.write('Learning Rate, Iterations, Batch Size, Five-Fold\n' + str(learning_rate) + ',' + str(iteration) +
+                    ',' + str(batch_size) + ',' + str(five_fold))
+        hyper.close()
+
         # Getting Train-data directory
         aug_data_path = os.getcwd() + "/aug_Data"
         if tf.gfile.Exists(aug_data_path):
@@ -84,7 +113,8 @@ def quit_button_applied():
                     weight_path = accessories_path + "/Five_Fold_Trained/Trained_Fold_" + str(num+1)
                     tf.gfile.MakeDirs(weight_path)
                     if tf.gfile.Exists(train_path):
-                        melNET_train_src.main(train_path, weight_path, learning_rate, batch_size, iteration)
+                        melNET_train_src.main(train_path, weight_path, float(learning_rate), int(batch_size),
+                                              int(iteration))
                     else:
                         print("Training Data is NOT available!")
             else:  # Single-Fold
@@ -93,7 +123,7 @@ def quit_button_applied():
                 weight_path = accessories_path + "/Single_Fold_Trained"
                 tf.gfile.MakeDirs(weight_path)
                 if tf.gfile.Exists(train_path):
-                    melNET_train_src.main(train_path, weight_path, learning_rate, batch_size, iteration)
+                    melNET_train_src.main(train_path, weight_path, float(learning_rate), int(batch_size), int(iteration))
                 else:
                     print("Training Data is NOT available!")
         else: