This is a fun project I worked on after completing the first course of the deeplearning specialization on coursera by deeplearning.ai. This is not an accurate model, I did a quick rough implementation of the binary classification model that I created for an assignment in the course.
# TensorFlow and tf.keras
import tensorflow as tf
from tensorflow import keras
# Helper libraries
import numpy as np
import matplotlib.pyplot as plt
from google.colab import drive
import cv2
from sklearn.model_selection import train_test_split
plt.rcParams['figure.figsize'] = (5.0, 4.0) # set default size of plots
plt.rcParams['image.interpolation'] = 'nearest'
plt.rcParams['image.cmap'] = 'gray'
import osdrive.mount('/content/gdrive/')Drive already mounted at /content/gdrive/; to attempt to forcibly remount, call drive.mount("/content/gdrive/", force_remount=True).
Get the list of image files located in the folder 'faces'
images = os.listdir('gdrive/My Drive/Colab Notebooks/faces')This is a function to only use the files with classes : anger, fear, disgust, happy, sad, surprised, and neutral. The pictures are from VISGRAF faces Database and Japanese Female Facial Expression (JAFFE) Database.
def contains_keyword(file):
keywords = ['AN', 'DI', 'FE', 'HA', 'SA', 'SU', 'NE', '-00','-01','-02','-03','-04','-05','-06']
b_list = [i in files for i in keywords]
if True in b_list:
return True
else:
return Falselabels = []
emotions = []
X = []
num_px = 256
# neutral = 0; anger = 1; disgust = 2; fear = 3; happy = 4; sad = 5; surprised = 6;
for files in images:
if contains_keyword(files):
fname = "gdrive/My Drive/Colab Notebooks/faces/" + files
img = cv2.imread(fname)
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
my_image = cv2.resize(img, (num_px, num_px))
my_image = img/255.
X.append(my_image)
if 'AN' in files or '-04' in files:
emotions.append('Angry')
labels.append(1)
elif 'DI' in files or '-05' in files:
emotions.append('Disgust')
labels.append(2)
elif 'FE' in files or '-06' in files:
emotions.append('Fear')
labels.append(3)
elif 'HA' in files or '-01' in files:
emotions.append('Happy')
labels.append(4)
elif 'SA' in files or '-02' in files:
emotions.append('Sad')
labels.append(5)
elif 'SU' in files or '-03' in files:
emotions.append('Surprised')
labels.append(6)
elif 'NE' in files or '-00' in files:
emotions.append('Nuetral')
labels.append(0)class_names = ['Neutral', 'Anger', 'Disgust', 'Fear', 'Happy',
'Sad', 'Surprised']# total data size
m = len(labels)
Y = np.array(labels)
X = np.array(X)
X.shape(453, 256, 256)
X_flatten = X.reshape(X.shape[0], -1)len(Y)453
plt.figure()
plt.imshow(X[0])
plt.colorbar()
plt.grid(False)
plt.show()
plt.figure(figsize=(10,10))
for i in range(25):
plt.subplot(5,5,i+1)
plt.xticks([])
plt.yticks([])
plt.grid(False)
plt.imshow(X[i])
plt.xlabel(class_names[Y[i]])
plt.show()
X_flatten[0].shape(65536,)
X_train, X_test, Y_train, Y_test = train_test_split(X_flatten, Y, test_size=0.20, random_state=42)X_train.shape(362, 65536)
Y_trainarray([1, 5, 0, 3, 6, 1, 6, 0, 6, 1, 6, 2, 5, 0, 0, 3, 6, 5, 6, 1, 5, 0,
3, 1, 2, 3, 1, 5, 0, 6, 0, 4, 5, 5, 4, 5, 4, 6, 0, 0, 1, 0, 1, 3,
3, 2, 5, 1, 3, 0, 4, 3, 5, 0, 5, 6, 2, 1, 2, 0, 6, 6, 4, 4, 0, 3,
4, 0, 2, 3, 4, 1, 2, 0, 1, 3, 6, 5, 5, 1, 3, 3, 0, 6, 4, 6, 3, 3,
2, 3, 3, 2, 1, 0, 6, 3, 2, 3, 4, 0, 4, 0, 1, 3, 6, 4, 5, 3, 4, 1,
4, 5, 0, 5, 0, 4, 5, 4, 4, 0, 2, 6, 0, 3, 0, 4, 3, 1, 0, 4, 0, 4,
5, 1, 4, 6, 3, 4, 1, 1, 2, 0, 5, 0, 6, 0, 3, 2, 1, 6, 6, 1, 3, 6,
5, 5, 2, 1, 2, 5, 6, 1, 0, 4, 6, 6, 4, 6, 1, 1, 2, 2, 2, 5, 4, 1,
6, 2, 2, 1, 1, 0, 4, 3, 5, 6, 0, 6, 5, 1, 2, 1, 5, 6, 6, 4, 6, 4,
0, 4, 6, 4, 0, 4, 0, 4, 5, 6, 1, 6, 1, 6, 3, 5, 4, 1, 6, 3, 2, 1,
3, 2, 6, 2, 3, 5, 1, 3, 4, 6, 3, 5, 0, 4, 2, 6, 3, 3, 5, 5, 6, 1,
4, 2, 1, 0, 3, 1, 4, 6, 1, 5, 3, 0, 2, 4, 4, 6, 6, 2, 3, 2, 2, 1,
1, 4, 2, 4, 2, 3, 1, 6, 0, 1, 4, 6, 3, 5, 3, 0, 6, 5, 2, 1, 2, 4,
6, 4, 5, 5, 4, 4, 1, 2, 2, 3, 5, 2, 4, 2, 5, 3, 3, 5, 6, 0, 0, 0,
6, 1, 1, 0, 2, 3, 3, 2, 5, 2, 6, 0, 5, 2, 4, 1, 0, 1, 6, 1, 2, 6,
4, 5, 4, 2, 1, 3, 2, 3, 6, 4, 1, 5, 5, 0, 5, 3, 3, 5, 4, 2, 0, 4,
1, 0, 4, 2, 0, 3, 0, 3, 2, 6])
model = keras.Sequential([
keras.layers.InputLayer(input_shape = (65536,)),
keras.layers.Dense(128, activation = 'relu'),
keras.layers.Dense(128, activation = 'relu'),
keras.layers.Dense(128, activation = 'relu'),
keras.layers.Dense(7, activity_regularizer=tf.keras.regularizers.l2(0.001))
])model.compile(optimizer='Adamax',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy'], run_eagerly=True)model.fit(X_train, Y_train, epochs=60)Epoch 1/60
12/12 [==============================] - 1s 65ms/step - loss: 8.0944 - accuracy: 0.1298
Epoch 2/60
12/12 [==============================] - 1s 61ms/step - loss: 2.6393 - accuracy: 0.2099
Epoch 3/60
12/12 [==============================] - 1s 62ms/step - loss: 2.0447 - accuracy: 0.2099
Epoch 4/60
12/12 [==============================] - 1s 61ms/step - loss: 2.0394 - accuracy: 0.2320
Epoch 5/60
12/12 [==============================] - 1s 60ms/step - loss: 2.0246 - accuracy: 0.2762
Epoch 6/60
12/12 [==============================] - 1s 61ms/step - loss: 1.8533 - accuracy: 0.2735
Epoch 7/60
12/12 [==============================] - 1s 62ms/step - loss: 1.7749 - accuracy: 0.2735
Epoch 8/60
12/12 [==============================] - 1s 61ms/step - loss: 1.8011 - accuracy: 0.4282
Epoch 9/60
12/12 [==============================] - 1s 60ms/step - loss: 1.9108 - accuracy: 0.2707
Epoch 10/60
12/12 [==============================] - 1s 61ms/step - loss: 1.7970 - accuracy: 0.2873
Epoch 11/60
12/12 [==============================] - 1s 59ms/step - loss: 1.6215 - accuracy: 0.3867
Epoch 12/60
12/12 [==============================] - 1s 59ms/step - loss: 1.4846 - accuracy: 0.4530
Epoch 13/60
12/12 [==============================] - 1s 60ms/step - loss: 1.4595 - accuracy: 0.4807
Epoch 14/60
12/12 [==============================] - 1s 62ms/step - loss: 1.4428 - accuracy: 0.4751
Epoch 15/60
12/12 [==============================] - 1s 60ms/step - loss: 1.4446 - accuracy: 0.4834
Epoch 16/60
12/12 [==============================] - 1s 62ms/step - loss: 1.4156 - accuracy: 0.4862
Epoch 17/60
12/12 [==============================] - 1s 61ms/step - loss: 1.3760 - accuracy: 0.4696
Epoch 18/60
12/12 [==============================] - 1s 63ms/step - loss: 1.2869 - accuracy: 0.5442
Epoch 19/60
12/12 [==============================] - 1s 60ms/step - loss: 1.2834 - accuracy: 0.5497
Epoch 20/60
12/12 [==============================] - 1s 62ms/step - loss: 1.3357 - accuracy: 0.5414
Epoch 21/60
12/12 [==============================] - 1s 60ms/step - loss: 1.3421 - accuracy: 0.5221
Epoch 22/60
12/12 [==============================] - 1s 60ms/step - loss: 1.1680 - accuracy: 0.6022
Epoch 23/60
12/12 [==============================] - 1s 59ms/step - loss: 1.2014 - accuracy: 0.6050
Epoch 24/60
12/12 [==============================] - 1s 61ms/step - loss: 1.1583 - accuracy: 0.5801
Epoch 25/60
12/12 [==============================] - 1s 60ms/step - loss: 1.1435 - accuracy: 0.6243
Epoch 26/60
12/12 [==============================] - 1s 59ms/step - loss: 1.0814 - accuracy: 0.6354
Epoch 27/60
12/12 [==============================] - 1s 60ms/step - loss: 1.0592 - accuracy: 0.6740
Epoch 28/60
12/12 [==============================] - 1s 59ms/step - loss: 1.1595 - accuracy: 0.5939
Epoch 29/60
12/12 [==============================] - 1s 58ms/step - loss: 1.0936 - accuracy: 0.6160
Epoch 30/60
12/12 [==============================] - 1s 60ms/step - loss: 1.1044 - accuracy: 0.6326
Epoch 31/60
12/12 [==============================] - 1s 59ms/step - loss: 1.0102 - accuracy: 0.6685
Epoch 32/60
12/12 [==============================] - 1s 58ms/step - loss: 0.9517 - accuracy: 0.7127
Epoch 33/60
12/12 [==============================] - 1s 59ms/step - loss: 0.9634 - accuracy: 0.7210
Epoch 34/60
12/12 [==============================] - 1s 59ms/step - loss: 0.9353 - accuracy: 0.7182
Epoch 35/60
12/12 [==============================] - 1s 59ms/step - loss: 0.9467 - accuracy: 0.7265
Epoch 36/60
12/12 [==============================] - 1s 59ms/step - loss: 0.9090 - accuracy: 0.7514
Epoch 37/60
12/12 [==============================] - 1s 59ms/step - loss: 0.8486 - accuracy: 0.7541
Epoch 38/60
12/12 [==============================] - 1s 59ms/step - loss: 0.9144 - accuracy: 0.7127
Epoch 39/60
12/12 [==============================] - 1s 60ms/step - loss: 0.8321 - accuracy: 0.7597
Epoch 40/60
12/12 [==============================] - 1s 59ms/step - loss: 0.7881 - accuracy: 0.7845
Epoch 41/60
12/12 [==============================] - 1s 60ms/step - loss: 0.7637 - accuracy: 0.7956
Epoch 42/60
12/12 [==============================] - 1s 59ms/step - loss: 0.7887 - accuracy: 0.8039
Epoch 43/60
12/12 [==============================] - 1s 62ms/step - loss: 0.7691 - accuracy: 0.7928
Epoch 44/60
12/12 [==============================] - 1s 60ms/step - loss: 0.7775 - accuracy: 0.7680
Epoch 45/60
12/12 [==============================] - 1s 61ms/step - loss: 0.7472 - accuracy: 0.8094
Epoch 46/60
12/12 [==============================] - 1s 61ms/step - loss: 0.7960 - accuracy: 0.7652
Epoch 47/60
12/12 [==============================] - 1s 60ms/step - loss: 0.7336 - accuracy: 0.7845
Epoch 48/60
12/12 [==============================] - 1s 60ms/step - loss: 0.8333 - accuracy: 0.7265
Epoch 49/60
12/12 [==============================] - 1s 59ms/step - loss: 0.7089 - accuracy: 0.7707
Epoch 50/60
12/12 [==============================] - 1s 60ms/step - loss: 0.6650 - accuracy: 0.7845
Epoch 51/60
12/12 [==============================] - 1s 60ms/step - loss: 0.6920 - accuracy: 0.8011
Epoch 52/60
12/12 [==============================] - 1s 60ms/step - loss: 0.6256 - accuracy: 0.8398
Epoch 53/60
12/12 [==============================] - 1s 59ms/step - loss: 0.7953 - accuracy: 0.7320
Epoch 54/60
12/12 [==============================] - 1s 59ms/step - loss: 0.6706 - accuracy: 0.7901
Epoch 55/60
12/12 [==============================] - 1s 61ms/step - loss: 0.6228 - accuracy: 0.8149
Epoch 56/60
12/12 [==============================] - 1s 63ms/step - loss: 0.5587 - accuracy: 0.8729
Epoch 57/60
12/12 [==============================] - 1s 60ms/step - loss: 0.5595 - accuracy: 0.8729
Epoch 58/60
12/12 [==============================] - 1s 59ms/step - loss: 0.4965 - accuracy: 0.9006
Epoch 59/60
12/12 [==============================] - 1s 60ms/step - loss: 0.5273 - accuracy: 0.8867
Epoch 60/60
12/12 [==============================] - 1s 61ms/step - loss: 0.5631 - accuracy: 0.8398
<tensorflow.python.keras.callbacks.History at 0x7f2891457a20>
est_loss, test_acc = model.evaluate(X_test, Y_test, verbose=2)
print('\nTest accuracy:', test_acc)3/3 - 0s - loss: 1.5569 - accuracy: 0.5275
Test accuracy: 0.5274725556373596
probability_model = tf.keras.Sequential([model,
tf.keras.layers.Softmax()])predictions = probability_model.predict(X_test)predictions[0]array([0.28819346, 0.07667705, 0.018339 , 0.04596484, 0.01074533,
0.06403587, 0.49604434], dtype=float32)
np.argmax(predictions[0])6
def plot_image(i, predictions_array, true_label, img):
predictions_array, true_label, img = predictions_array, true_label[i], img[i]
plt.grid(False)
plt.xticks([])
plt.yticks([])
plt.imshow(img, cmap='gray')
predicted_label = np.argmax(predictions_array)
if predicted_label == true_label:
color = 'blue'
else:
color = 'red'
plt.xlabel("{} {:2.0f}% ({})".format(class_names[predicted_label],
100*np.max(predictions_array),
class_names[true_label]),
color=color)
def plot_value_array(i, predictions_array, true_label):
predictions_array, true_label = predictions_array, true_label[i]
plt.grid(False)
plt.xticks(range(7))
plt.yticks([])
thisplot = plt.bar(range(7), predictions_array, color="#777777")
plt.ylim([0, 1])
predicted_label = np.argmax(predictions_array)
thisplot[predicted_label].set_color('red')
thisplot[true_label].set_color('blue')The following code will show you an image in the dataset. Feel free to change the index and re-run the cell multiple times to see other images.
X_test.reshape(91, 256,256).shape(91, 256, 256)
X_test.shape
Y_test.shape(91,)
i = 0
plt.figure(figsize=(6,3))
plt.subplot(1,2,1)
plot_image(i, predictions[i], Y_test, X_test.reshape(91, 256,256))
plt.subplot(1,2,2)
plot_value_array(i, predictions[i], Y_test)
plt.show()
i = 65
plt.figure(figsize=(6,3))
plt.subplot(1,2,1)
plot_image(i, predictions[i], Y_test, X_test.reshape(91, 256,256))
plt.subplot(1,2,2)
plot_value_array(i, predictions[i], Y_test)
plt.show()
num_rows = 5
num_cols = 3
num_images = num_rows*num_cols
plt.figure(figsize=(2*2*num_cols, 2*num_rows))
for i in range(num_images):
plt.subplot(num_rows, 2*num_cols, 2*i+1)
plot_image(i, predictions[i], Y_test, X_test.reshape(91,256,256))
plt.subplot(num_rows, 2*num_cols, 2*i+2)
plot_value_array(i, predictions[i], Y_test)
plt.tight_layout()
plt.show()