Directory restructuring

Author: pratikgujjar

code/autoencoder_model/scripts/.res_16.py.swp

@@ -75,6 +75,7 @@
 print ("Config file: " + str(__name__))
 
 BATCH_SIZE = 9
+TEST_BATCH_SIZE = 1
 NB_EPOCHS_AUTOENCODER = 30
 
 # OPTIM_A = Adam(lr=0.0001, beta_1=0.5)

code/autoencoder_model/scripts/config_r16.py

@@ -12,7 +12,7 @@
 import cv2
 import ast
 import os
-from config_aa import TEST_RESULTS_DIR
+from config_r16 import TEST_RESULTS_DIR
 
 def plot_heatmap(attn_layer, epoch, vid_num, file):
     print (attn_layer)
@@ -40,6 +40,14 @@ def plot_heatmap(attn_layer, epoch, vid_num, file):
                             transparent=True)
 
 
+def plot_err_variation(values, index):
+    plt.clf()
+    plt.plot(values)
+    plt.axis('off')
+    plt.savefig(os.path.join(TEST_RESULTS_DIR + '/graphs/', 'plot_' + str(index) + '.png'),
+                transparent=True)
+
+
 def get_args():
     parser = argparse.ArgumentParser()
     parser.add_argument("--file", type=str, default="None")

code/autoencoder_model/scripts/plot_heatmap.py

@@ -22,6 +22,8 @@
 from keras.layers.normalization import BatchNormalization
 from keras.callbacks import LearningRateScheduler
 from keras.layers.advanced_activations import LeakyReLU
+from sklearn.metrics import mean_absolute_error as mae
+from plot_results import plot_err_variation
 from keras.layers import Input
 from keras.models import Model
 from config_r16 import *
@@ -285,7 +287,7 @@ def load_to_RAM(frames_source):
         im_file = os.path.join(DATA_DIR, filename)
         try:
             frame = cv2.imread(im_file, cv2.IMREAD_COLOR)
-            # frame = cv2.resize(frame, (112, 112), interpolation=cv2.INTER_CUBIC)
+            frame = cv2.medianBlur(frame, FILTER_SIZE)
             frames[i] = (frame.astype(np.float32) - 127.5) / 127.5
             j = j + 1
         except AttributeError as e:
@@ -314,7 +316,7 @@ def load_X(videos_list, index, data_dir, img_size):
             im_file = os.path.join(data_dir, filename)
             try:
                 frame = cv2.imread(im_file, cv2.IMREAD_COLOR)
-                # frame = cv2.resize(frame, (112, 112), interpolation=cv2.INTER_LANCZOS4)
+                frame = cv2.medianBlur(frame, FILTER_SIZE)
                 X[i, j] = (frame.astype(np.float32) - 127.5) / 127.5
             except AttributeError as e:
                 print (im_file)
@@ -345,8 +347,8 @@ def get_video_lists(frames_source, stride):
 
 def train(BATCH_SIZE, ENC_WEIGHTS, DEC_WEIGHTS):
     print ("Loading data definitions...")
-    frames_source = hkl.load(os.path.join(DATA_DIR, 'sources_val_208.hkl'))
-    videos_list = get_video_lists(frames_source=frames_source, stride=8)
+    frames_source = hkl.load(os.path.join(DATA_DIR, 'sources_train_208.hkl'))
+    videos_list = get_video_lists(frames_source=frames_source, stride=4)
     n_videos = videos_list.shape[0]
 
     # Setup test
@@ -443,7 +445,7 @@ def train(BATCH_SIZE, ENC_WEIGHTS, DEC_WEIGHTS):
         # then after each epoch/iteration
         avg_loss = sum(loss)/len(loss)
         avg_val_loss = sum(val_loss) / len(val_loss)
-        logs = {'loss': avg_loss, 'test_loss': avg_val_loss}
+        logs = {'loss': avg_loss, 'val_loss': avg_val_loss}
         TC.on_epoch_end(epoch, logs)
 
         # Log the losses
@@ -477,8 +479,9 @@ def test(ENC_WEIGHTS, DEC_WEIGHTS):
         os.mkdir(TEST_RESULTS_DIR + '/pred/')
     if not os.path.exists(TEST_RESULTS_DIR + '/graphs/'):
         os.mkdir(TEST_RESULTS_DIR + '/graphs/')
-    if not os.path.exists(TEST_RESULTS_DIR + '/gifs/'):
-        os.mkdir(TEST_RESULTS_DIR + '/gifs/')
+        os.mkdir(TEST_RESULTS_DIR + '/graphs/values/')
+    # if not os.path.exists(TEST_RESULTS_DIR + '/gifs/'):
+    #     os.mkdir(TEST_RESULTS_DIR + '/gifs/')
 
     print("Creating models...")
     encoder = encoder_model()
@@ -489,7 +492,7 @@ def test(ENC_WEIGHTS, DEC_WEIGHTS):
 
     run_utilities(encoder, decoder, autoencoder, ENC_WEIGHTS, DEC_WEIGHTS)
 
-    NB_TEST_ITERATIONS = int(n_test_videos / BATCH_SIZE)
+    NB_TEST_ITERATIONS = int(n_test_videos / TEST_BATCH_SIZE)
     test_loss = []
     for index in range(NB_TEST_ITERATIONS):
         X = load_X(test_videos_list, index, TEST_DATA_DIR, IMG_SIZE)
@@ -513,8 +516,15 @@ def test(ENC_WEIGHTS, DEC_WEIGHTS):
             truth_seq = truth_seq * 127.5 + 127.5
             pred_seq = pred_seq * 127.5 + 127.5
 
+            mae_errors = []
+            for i in range(int(VIDEO_LENGTH / 2)):
+                mae_errors.append(mae(y_test[i].flatten(), predicted_images[i].flatten()))
+
             cv2.imwrite(os.path.join(TEST_RESULTS_DIR + '/truth/', str(index) + "_truth.png"), truth_seq)
             cv2.imwrite(os.path.join(TEST_RESULTS_DIR + '/pred/', str(index) + "_pred.png"), pred_seq)
+            np.save(os.path.join(TEST_RESULTS_DIR + '/graphs/values/', str(index) + "_mae.npy"), np.asarray(mae_errors))
+            plot_err_variation(mae_errors, index)
+
 
     # then after each epoch/iteration
     avg_test_loss = sum(test_loss) / len(test_loss)
@@ -527,10 +537,6 @@ def test(ENC_WEIGHTS, DEC_WEIGHTS):
     print("\n Min: " + str(np.min(np.asarray(test_loss))))
 
 
-def err_variation()
-    K.abs(y_pred - y_true)
-
-
 def load_X_test(index, data_dir, img_size):
     X = np.zeros((BATCH_SIZE, int(VIDEO_LENGTH/2),) + img_size)
     for i in range(BATCH_SIZE):

code/autoencoder_model/scripts/res_16.py

@@ -0,0 +1,102 @@
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from keras.optimizers import SGD
+from keras.optimizers import Adam
+from keras.optimizers import adadelta
+from keras.optimizers import rmsprop
+from keras.layers import Layer
+from keras import backend as K
+K.set_image_dim_ordering('tf')
+import socket
+import os
+
+# -------------------------------------------------
+# Background config:
+hostname = socket.gethostname()
+if hostname == 'baymax':
+    path_var = 'baymax/'
+elif hostname == 'walle':
+    path_var = 'walle/'
+elif hostname == 'bender':
+    path_var = 'bender/'
+else:
+    path_var = 'zhora/'
+
+DATA_DIR= '/local_home/JAAD_Dataset/iros/resized_imgs_208_thesis/train/'
+
+VAL_DATA_DIR= '/local_home/JAAD_Dataset/iros/resized_imgs_208_thesis/val/'
+
+TEST_DATA_DIR= '/local_home/JAAD_Dataset/iros/resized_imgs_208_thesis/test/'
+# TEST_DATA_DIR= '/local_home/JAAD_Dataset/fun_experiments/resized/'
+
+RESULTS_DIR = '/local_home/JAAD_Dataset/thesis/results/DRRK16/'
+
+# MODEL_DIR = './../' + path_var + 'models'
+MODEL_DIR = RESULTS_DIR + 'models/'
+if not os.path.exists(MODEL_DIR):
+    os.mkdir(MODEL_DIR)
+
+# CHECKPOINT_DIR = './../' + path_var + 'checkpoints'
+CHECKPOINT_DIR = RESULTS_DIR + 'checkpoints/'
+if not os.path.exists(CHECKPOINT_DIR):
+    os.mkdir(CHECKPOINT_DIR)
+
+GEN_IMAGES_DIR = RESULTS_DIR + 'generated_images/'
+if not os.path.exists(GEN_IMAGES_DIR):
+    os.mkdir(GEN_IMAGES_DIR)
+
+LOG_DIR = RESULTS_DIR + 'logs/'
+if not os.path.exists(LOG_DIR):
+    os.mkdir(LOG_DIR)
+
+TF_LOG_DIR = RESULTS_DIR + 'tf_logs/'
+if not os.path.exists(TF_LOG_DIR):
+    os.mkdir(TF_LOG_DIR)
+
+TEST_RESULTS_DIR = RESULTS_DIR + 'test_results/'
+if not os.path.exists(TEST_RESULTS_DIR):
+    os.mkdir(TEST_RESULTS_DIR)
+
+PRINT_MODEL_SUMMARY = True
+SAVE_MODEL = True
+PLOT_MODEL = True
+SAVE_GENERATED_IMAGES = True
+SHUFFLE = True
+VIDEO_LENGTH = 32
+IMG_SIZE = (128, 208, 3)
+RAM_DECIMATE = True
+REVERSE = True
+FILTER_SIZE = 3
+
+
+# -------------------------------------------------
+# Network configuration:
+print ("Loading network/training configuration.")
+print ("Config file: " + str(__name__))
+
+BATCH_SIZE = 9
+TEST_BATCH_SIZE = 1
+NB_EPOCHS_AUTOENCODER = 30
+
+# OPTIM_A = Adam(lr=0.0001, beta_1=0.5)
+OPTIM_A = rmsprop(lr=0.0001, rho=0.9)
+OPTIM_B = rmsprop(lr=0.00001, rho=0.9)
+# OPTIM_A = SGD(lr=0.000001, momentum=0.5, nesterov=True)
+
+lr_schedule = [7, 14, 20, 30]  # epoch_step
+
+def schedule(epoch_idx):
+    if (epoch_idx) <= lr_schedule[0]:
+        return 0.001
+    elif (epoch_idx) <= lr_schedule[1]:
+        return 0.0001  # lr_decay_ratio = 10
+    elif (epoch_idx) <= lr_schedule[2]:
+        return 0.00001  # lr_decay_ratio = 10
+    elif (epoch_idx) <= lr_schedule[3]:
+        return 0.00001
+    return 0.00001
+
+
+