Write code to calculate MTCP

Author: pratikgujjar

code/autoencoder_model/scripts/thesis_scripts/classifier_all.py

@@ -300,7 +300,10 @@ def load_to_RAM(frames_source):
         im_file = os.path.join(DATA_DIR, filename)
         try:
             frame = cv2.imread(im_file, cv2.IMREAD_COLOR)
-            frames[i] = frame.astype(np.float32)
+            if RANDOM_AUGMENTATION:
+                frames[i] = frame.astype(np.float32)
+            else:
+                frames[i] = (frame.astype(np.float32) - 127.5) / 127.5
             j = j + 1
         except AttributeError as e:
             print(im_file)
@@ -609,7 +612,7 @@ def train(BATCH_SIZE, ENC_WEIGHTS, DEC_WEIGHTS, CLA_WEIGHTS):
     # Load test action annotations
     val_action_labels = hkl.load(os.path.join(VAL_DATA_DIR, 'annotations_val_208.hkl'))
     val_ped_action_classes, val_ped_class_count = get_action_classes(val_action_labels)
-    print("VAl Stats: " + str(val_ped_class_count))
+    print("Val Stats: " + str(val_ped_class_count))
 
     # Build the Spatio-temporal Autoencoder
     print ("Creating models.")
@@ -670,7 +673,7 @@ def train(BATCH_SIZE, ENC_WEIGHTS, DEC_WEIGHTS, CLA_WEIGHTS):
             c_loss.append(sclassifier.train_on_batch(X_train, [y_true_imgs, y_true_class]))
 
             y_train_true.extend(y_true_class)
-            y_train_pred.extend(sclassifier.predict(X_train, verbose=0))
+            y_train_pred.extend(sclassifier.predict(X_train, verbose=0)[1])
 
             arrow = int(index / (NB_ITERATIONS / 30))
             stdout.write("\rIter: " + str(index) + "/" + str(NB_ITERATIONS - 1) + "  " +
@@ -738,7 +741,7 @@ def train(BATCH_SIZE, ENC_WEIGHTS, DEC_WEIGHTS, CLA_WEIGHTS):
 
             val_c_loss.append(sclassifier.test_on_batch(X_val, [y_true_imgs, y_true_class]))
             y_val_true.extend(y_true_class)
-            y_val_pred.extend(sclassifier.predict(X_val, verbose=0))
+            y_val_pred.extend(sclassifier.predict(X_val, verbose=0)[1])
 
             arrow = int(index / (NB_VAL_ITERATIONS / 40))
             stdout.write("\rIter: " + str(index) + "/" + str(NB_VAL_ITERATIONS - 1) + "  " +
@@ -812,16 +815,16 @@ def train(BATCH_SIZE, ENC_WEIGHTS, DEC_WEIGHTS, CLA_WEIGHTS):
                                                                             avg='binary',
                                                                             pos_label=1)
 
-        print("Train Prec: %.2f, Recall: %.2f, Fbeta: %.2f" % (train_prec, train_rec, train_fbeta))
+        print("\nTrain Prec: %.2f, Recall: %.2f, Fbeta: %.2f" % (train_prec, train_rec, train_fbeta))
         print("Val Prec: %.2f, Recall: %.2f, Fbeta: %.2f" % (val_prec, val_rec, val_fbeta))
-        # loss_values = np.asarray(avg_c_loss.tolist() + [train_prec.tolist()] +
-        #                          [train_rec.tolist()] +
-        #                          avg_val_c_loss.tolist() + [val_prec.tolist()] +
-        #                          [val_rec.tolist()], dtype=np.float32)
-        loss_values = np.asarray(avg_c_loss.tolist() + train_prec.tolist() +
-                                 train_rec.tolist() +
-                                 avg_val_c_loss.tolist() + val_prec.tolist() +
-                                 val_rec.tolist(), dtype=np.float32)
+        loss_values = np.asarray(avg_c_loss.tolist() + [train_prec.tolist()] +
+                                 [train_rec.tolist()] +
+                                 avg_val_c_loss.tolist() + [val_prec.tolist()] +
+                                 [val_rec.tolist()], dtype=np.float32)
+        # loss_values = np.asarray(avg_c_loss.tolist() + train_prec.tolist() +
+        #                          train_rec.tolist() +
+        #                          avg_val_c_loss.tolist() + val_prec.tolist() +
+        #                          val_rec.tolist(), dtype=np.float32)
 
         precs = ['prec_' + action for action in simple_ped_set]
         recs = ['rec_' + action for action in simple_ped_set]