Artificial Intelligence Nanodegree

Convolutional Neural Networks

Project: Write an Algorithm for a Dog Identification App


In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this IPython notebook.


Why We're Here

In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!

The Road Ahead

We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.

  • Step 0: Import Datasets
  • Step 1: Detect Humans
  • Step 2: Detect Dogs
  • Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
  • Step 4: Use a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 6: Write your Algorithm
  • Step 7: Test Your Algorithm

Step 0: Import Datasets

Import Dog Dataset

In the code cell below, we import a dataset of dog images. We populate a few variables through the use of the load_files function from the scikit-learn library:

  • train_files, valid_files, test_files - numpy arrays containing file paths to images
  • train_targets, valid_targets, test_targets - numpy arrays containing onehot-encoded classification labels
  • dog_names - list of string-valued dog breed names for translating labels

In [1]:
from sklearn.datasets import load_files       
from keras.utils import np_utils
import numpy as np
from glob import glob

# define function to load train, test, and validation datasets
def load_dataset(path):
    data = load_files(path)
    dog_files = np.array(data['filenames'])
    dog_targets = np_utils.to_categorical(np.array(data['target']), 133)
    return dog_files, dog_targets

# load train, test, and validation datasets
train_files, train_targets = load_dataset('/data/dog_images/train')
valid_files, valid_targets = load_dataset('/data/dog_images/valid')
test_files, test_targets = load_dataset('/data/dog_images/test')

# load list of dog names
dog_names = [item[20:-1] for item in sorted(glob("/data/dog_images/train/*/"))]

# print statistics about the dataset
print('There are %d total dog categories.' % len(dog_names))
print('There are %s total dog images.\n' % len(np.hstack([train_files, valid_files, test_files])))
print('There are %d training dog images.' % len(train_files))
print('There are %d validation dog images.' % len(valid_files))
print('There are %d test dog images.'% len(test_files))


Using TensorFlow backend.
There are 133 total dog categories.
There are 8351 total dog images.

There are 6680 training dog images.
There are 835 validation dog images.
There are 836 test dog images.

Import Human Dataset

In the code cell below, we import a dataset of human images, where the file paths are stored in the numpy array human_files.


In [2]:
import random
random.seed(8675309)

# load filenames in shuffled human dataset
human_files = np.array(glob("/data/lfw/*/*"))
random.shuffle(human_files)

# print statistics about the dataset
print('There are %d total human images.' % len(human_files))


There are 13233 total human images.

Step 1: Detect Humans

We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory.

In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.


In [3]:
import cv2                
import matplotlib.pyplot as plt                        
%matplotlib inline                               

# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')

# load color (BGR) image
img = cv2.imread(human_files[3])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# find faces in image
faces = face_cascade.detectMultiScale(gray)

# print number of faces detected in the image
print('Number of faces detected:', len(faces))

# get bounding box for each detected face
for (x,y,w,h) in faces:
    # add bounding box to color image
    cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
    
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()


Number of faces detected: 1

Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.

Write a Human Face Detector

We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.


In [4]:
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
    return len(faces) > 0

(IMPLEMENTATION) Assess the Human Face Detector

Question 1: Use the code cell below to test the performance of the face_detector function.

  • What percentage of the first 100 images in human_files have a detected human face?
  • What percentage of the first 100 images in dog_files have a detected human face?

Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.

Answer:

  • Of the first 100 images in 'human_files', the face detector detects a human face in each of the 100 files, hence the accuracy is 1.0 and 100% of the images have a human face detected in them.
  • This detector has been trained on human faces and sometimes misclassifies dog faces as human faces. It does so for 11 of the first 100 images in 'dog_files'. 11% of the dog images has a human face detected in them.

In [5]:
human_files_short = human_files[:100]
dog_files_short = train_files[:100]
# Do NOT modify the code above this line.

## TODO: Test the performance of the face_detector algorithm 
## on the images in human_files_short and dog_files_short.

humans = 0
dogs = 0

for img in human_files_short:
    if face_detector(img) :
        humans = humans + 1

for img in dog_files_short:
    if face_detector(img) :
        dogs = dogs + 1
        
accuracy_on_humans = humans / 100.0        
accuracy_on_dogs = dogs / 100.0

print ("Accuracy on humans is " + str(accuracy_on_humans))
print ("Accuracy on dogs is " + str(accuracy_on_dogs))


Accuracy on humans is 1.0
Accuracy on dogs is 0.11

Question 2: This algorithmic choice necessitates that we communicate to the user that we accept human images only when they provide a clear view of a face (otherwise, we risk having unneccessarily frustrated users!). In your opinion, is this a reasonable expectation to pose on the user? If not, can you think of a way to detect humans in images that does not necessitate an image with a clearly presented face?

Answer: Asking a user to provide a clear view of the human face is not a great experience for the user. It would be better if we could associate a confidence interval with our predictions and use that to choose an operational point that balances our precision and recall. Adding non-human faces to the dataset, specially those with similar facial features like eyes, human like ears and using them to retrain the algorithm to correctly identify humans in an image is a much preferred way.

We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on each of the datasets.

Haar Cascades Haar like features are named after Haar-wavelets. The wavelet has a peculaiar form, where by it is of value 1 for the range [0,1/2] and -1 for the range [-1/2,1]. The use of wavelets allows the images to be processed in the single dimension of intensities rather than using individual channels such as RGB pixel values.

As the Haar wavelet is discontinuous it is not differentiable, this property is advantageous when detecting edges and sudden transitions. This makes it suitable for dtecting objects within the images. Haar like features could also be used for dtecting faces in images. A Haar-like feature considers adjacent regions and sums up the pixels within a region. It then calculates the difference between the sums of adjacent rectangles. This difference is used to categorize subsections of an image. For a human face the region near the eye are known to be darker than the region near the cheeks. Therefore Haar like features are common for creating weak classifiers for detection of faces. In the work by Viola-Jones, a rectangular window of target size is moved over the entire image and for each subsection Haar like features are computed. A large number of such Haar-like features provide a set of classifiers that are organized into a hieracrchy known as Haar-cascade. Such a strong classifier is able to detect human faces with a higher accuracy.

Haar cascades are definitely better than a single classifier as they exploit they adaptively adjust the window sizes and are more generalizable.


In [6]:
## (Optional) TODO: Report the performance of another  
## face detection algorithm on the LFW dataset
### Feel free to use as many code cells as needed.

Step 2: Detect Dogs

In this section, we use a pre-trained ResNet-50 model to detect dogs in images. Our first line of code downloads the ResNet-50 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories. Given an image, this pre-trained ResNet-50 model returns a prediction (derived from the available categories in ImageNet) for the object that is contained in the image.


In [7]:
from keras.applications.resnet50 import ResNet50

# define ResNet50 model
ResNet50_model = ResNet50(weights='imagenet')


Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.2/resnet50_weights_tf_dim_ordering_tf_kernels.h5
102858752/102853048 [==============================] - 1s 0us/step

Pre-process the Data

When using TensorFlow as backend, Keras CNNs require a 4D array (which we'll also refer to as a 4D tensor) as input, with shape

$$ (\text{nb_samples}, \text{rows}, \text{columns}, \text{channels}), $$

where nb_samples corresponds to the total number of images (or samples), and rows, columns, and channels correspond to the number of rows, columns, and channels for each image, respectively.

The path_to_tensor function below takes a string-valued file path to a color image as input and returns a 4D tensor suitable for supplying to a Keras CNN. The function first loads the image and resizes it to a square image that is $224 \times 224$ pixels. Next, the image is converted to an array, which is then resized to a 4D tensor. In this case, since we are working with color images, each image has three channels. Likewise, since we are processing a single image (or sample), the returned tensor will always have shape

$$ (1, 224, 224, 3). $$

The paths_to_tensor function takes a numpy array of string-valued image paths as input and returns a 4D tensor with shape

$$ (\text{nb_samples}, 224, 224, 3). $$

Here, nb_samples is the number of samples, or number of images, in the supplied array of image paths. It is best to think of nb_samples as the number of 3D tensors (where each 3D tensor corresponds to a different image) in your dataset!


In [8]:
from keras.preprocessing import image                  
from tqdm import tqdm

def path_to_tensor(img_path):
    # loads RGB image as PIL.Image.Image type
    img = image.load_img(img_path, target_size=(224, 224))
    # convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
    x = image.img_to_array(img)
    # convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
    return np.expand_dims(x, axis=0)

def paths_to_tensor(img_paths):
    list_of_tensors = [path_to_tensor(img_path) for img_path in tqdm(img_paths)]
    return np.vstack(list_of_tensors)

Making Predictions with ResNet-50

Getting the 4D tensor ready for ResNet-50, and for any other pre-trained model in Keras, requires some additional processing. First, the RGB image is converted to BGR by reordering the channels. All pre-trained models have the additional normalization step that the mean pixel (expressed in RGB as $[103.939, 116.779, 123.68]$ and calculated from all pixels in all images in ImageNet) must be subtracted from every pixel in each image. This is implemented in the imported function preprocess_input. If you're curious, you can check the code for preprocess_input here.

Now that we have a way to format our image for supplying to ResNet-50, we are now ready to use the model to extract the predictions. This is accomplished with the predict method, which returns an array whose $i$-th entry is the model's predicted probability that the image belongs to the $i$-th ImageNet category. This is implemented in the ResNet50_predict_labels function below.

By taking the argmax of the predicted probability vector, we obtain an integer corresponding to the model's predicted object class, which we can identify with an object category through the use of this dictionary.


In [9]:
from keras.applications.resnet50 import preprocess_input, decode_predictions

def ResNet50_predict_labels(img_path):
    # returns prediction vector for image located at img_path
    img = preprocess_input(path_to_tensor(img_path))
    return np.argmax(ResNet50_model.predict(img))

Write a Dog Detector

While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained ResNet-50 model, we need only check if the ResNet50_predict_labels function above returns a value between 151 and 268 (inclusive).

We use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).


In [10]:
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
    prediction = ResNet50_predict_labels(img_path)
    return ((prediction <= 268) & (prediction >= 151))

(IMPLEMENTATION) Assess the Dog Detector

Question 3: Use the code cell below to test the performance of your dog_detector function.

  • What percentage of the images in human_files_short have a detected dog?
  • What percentage of the images in dog_files_short have a detected dog?

Answer:

  • 0% of the human images has a dog detected in them
  • 100% of dog imgaes have a dog detected in them

In [11]:
### TODO: Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.
humans = 0
dogs = 0

for img in human_files_short:
    if dog_detector(img) :
        humans = humans + 1

for img in dog_files_short:
    if dog_detector(img) :
        dogs = dogs + 1
        
accuracy_on_humans = humans / 100.0        
accuracy_on_dogs = dogs / 100.0

print ("Accuracy on humans is " + str(accuracy_on_humans))
print ("Accuracy on dogs is " + str(accuracy_on_dogs))


Accuracy on humans is 0.0
Accuracy on dogs is 1.0

Step 3: Create a CNN to Classify Dog Breeds (from Scratch)

Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 1%. In Step 5 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.

Be careful with adding too many trainable layers! More parameters means longer training, which means you are more likely to need a GPU to accelerate the training process. Thankfully, Keras provides a handy estimate of the time that each epoch is likely to take; you can extrapolate this estimate to figure out how long it will take for your algorithm to train.

We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have great difficulty in distinguishing between a Brittany and a Welsh Springer Spaniel.

Brittany Welsh Springer Spaniel

It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).

Curly-Coated Retriever American Water Spaniel

Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.

Yellow Labrador Chocolate Labrador Black Labrador

We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.

Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!

Pre-process the Data

We rescale the images by dividing every pixel in every image by 255.


In [12]:
from PIL import ImageFile                            
ImageFile.LOAD_TRUNCATED_IMAGES = True                 

# pre-process the data for Keras
train_tensors = paths_to_tensor(train_files).astype('float32')/255
valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
test_tensors = paths_to_tensor(test_files).astype('float32')/255


100%|██████████| 6680/6680 [01:13<00:00, 90.95it/s] 
100%|██████████| 835/835 [00:08<00:00, 100.34it/s]
100%|██████████| 836/836 [00:08<00:00, 101.32it/s]

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    model.summary()

We have imported some Python modules to get you started, but feel free to import as many modules as you need. If you end up getting stuck, here's a hint that specifies a model that trains relatively fast on CPU and attains >1% test accuracy in 5 epochs:

Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. If you chose to use the hinted architecture above, describe why you think that CNN architecture should work well for the image classification task.

Answer: I used the architecture hinted above, the three interleaved convolution and max pooling layer filters and aggregation layers that are able to capture the features such as edges, curves and the color contrast information in the image, which are helpful in learning distinguishing characteristsics for dog breeds.


In [13]:
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from keras.layers import Dropout, Flatten, Dense, Activation
from keras.models import Sequential

model = Sequential()

### TODO: Define your architecture.
# 1st Layer - Add an input layer of 32 nodes with the same input shape as
# the training samples in X
model.add(Conv2D(filters=16, kernel_size=2, padding='same', activation='relu', 
                        input_shape=(224, 224, 3)))

model.add(MaxPooling2D(pool_size=2))

model.add(Conv2D(filters=32, kernel_size=2, padding='same', activation='relu', 
                        input_shape=(110, 110, 3)))
model.add(MaxPooling2D(pool_size=2))

model.add(Conv2D(filters=64, kernel_size=2, padding='same', activation='relu', 
                        input_shape=(54, 54, 3)))
model.add(MaxPooling2D(pool_size=2))

model.add(GlobalAveragePooling2D())
model.add(Dense(133, activation='softmax'))

model.summary()


_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 224, 224, 16)      208       
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 112, 112, 16)      0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 112, 112, 32)      2080      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 56, 56, 32)        0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 56, 56, 64)        8256      
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 28, 28, 64)        0         
_________________________________________________________________
global_average_pooling2d_1 ( (None, 64)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 133)               8645      
=================================================================
Total params: 19,189
Trainable params: 19,189
Non-trainable params: 0
_________________________________________________________________

Compile the Model


In [14]:
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.


In [15]:
from keras.callbacks import ModelCheckpoint  

### TODO: specify the number of epochs that you would like to use to train the model.

epochs = 20

### Do NOT modify the code below this line.

checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch.hdf5', 
                               verbose=1, save_best_only=True)

model.fit(train_tensors, train_targets, 
          validation_data=(valid_tensors, valid_targets),
          epochs=epochs, batch_size=20, callbacks=[checkpointer], verbose=1)


Train on 6680 samples, validate on 835 samples
Epoch 1/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.8848 - acc: 0.0095Epoch 00001: val_loss improved from inf to 4.87154, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 22s 3ms/step - loss: 4.8847 - acc: 0.0094 - val_loss: 4.8715 - val_acc: 0.0108
Epoch 2/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.8697 - acc: 0.0116Epoch 00002: val_loss improved from 4.87154 to 4.86338, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 21s 3ms/step - loss: 4.8700 - acc: 0.0115 - val_loss: 4.8634 - val_acc: 0.0108
Epoch 3/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.8560 - acc: 0.0138Epoch 00003: val_loss improved from 4.86338 to 4.84458, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 22s 3ms/step - loss: 4.8556 - acc: 0.0138 - val_loss: 4.8446 - val_acc: 0.0156
Epoch 4/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.8131 - acc: 0.0156Epoch 00004: val_loss improved from 4.84458 to 4.80871, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 22s 3ms/step - loss: 4.8135 - acc: 0.0156 - val_loss: 4.8087 - val_acc: 0.0216
Epoch 5/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.7661 - acc: 0.0198Epoch 00005: val_loss improved from 4.80871 to 4.76712, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 22s 3ms/step - loss: 4.7663 - acc: 0.0199 - val_loss: 4.7671 - val_acc: 0.0275
Epoch 6/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.7261 - acc: 0.0266Epoch 00006: val_loss improved from 4.76712 to 4.74313, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 21s 3ms/step - loss: 4.7262 - acc: 0.0265 - val_loss: 4.7431 - val_acc: 0.0240
Epoch 7/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.6895 - acc: 0.0287Epoch 00007: val_loss improved from 4.74313 to 4.73179, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 21s 3ms/step - loss: 4.6892 - acc: 0.0286 - val_loss: 4.7318 - val_acc: 0.0275
Epoch 8/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.6549 - acc: 0.0330Epoch 00008: val_loss improved from 4.73179 to 4.69593, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 22s 3ms/step - loss: 4.6554 - acc: 0.0331 - val_loss: 4.6959 - val_acc: 0.0240
Epoch 9/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.6226 - acc: 0.0324Epoch 00009: val_loss did not improve
6680/6680 [==============================] - 22s 3ms/step - loss: 4.6221 - acc: 0.0323 - val_loss: 4.7332 - val_acc: 0.0251
Epoch 10/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.5962 - acc: 0.0380Epoch 00010: val_loss improved from 4.69593 to 4.67320, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 21s 3ms/step - loss: 4.5966 - acc: 0.0379 - val_loss: 4.6732 - val_acc: 0.0263
Epoch 11/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.5730 - acc: 0.0377Epoch 00011: val_loss improved from 4.67320 to 4.64045, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 22s 3ms/step - loss: 4.5730 - acc: 0.0379 - val_loss: 4.6405 - val_acc: 0.0347
Epoch 12/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.5520 - acc: 0.0407Epoch 00012: val_loss improved from 4.64045 to 4.63304, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 22s 3ms/step - loss: 4.5511 - acc: 0.0409 - val_loss: 4.6330 - val_acc: 0.0371
Epoch 13/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.5260 - acc: 0.0414Epoch 00013: val_loss did not improve
6680/6680 [==============================] - 22s 3ms/step - loss: 4.5258 - acc: 0.0416 - val_loss: 4.6732 - val_acc: 0.0359
Epoch 14/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.5125 - acc: 0.0470Epoch 00014: val_loss improved from 4.63304 to 4.61042, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 21s 3ms/step - loss: 4.5120 - acc: 0.0469 - val_loss: 4.6104 - val_acc: 0.0395
Epoch 15/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.4918 - acc: 0.0497Epoch 00015: val_loss improved from 4.61042 to 4.60525, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 21s 3ms/step - loss: 4.4917 - acc: 0.0497 - val_loss: 4.6053 - val_acc: 0.0431
Epoch 16/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.4755 - acc: 0.0514Epoch 00016: val_loss did not improve
6680/6680 [==============================] - 21s 3ms/step - loss: 4.4750 - acc: 0.0513 - val_loss: 4.6115 - val_acc: 0.0383
Epoch 17/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.4569 - acc: 0.0527Epoch 00017: val_loss improved from 4.60525 to 4.57103, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 21s 3ms/step - loss: 4.4569 - acc: 0.0528 - val_loss: 4.5710 - val_acc: 0.0395
Epoch 18/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.4422 - acc: 0.0548Epoch 00018: val_loss improved from 4.57103 to 4.54191, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 21s 3ms/step - loss: 4.4414 - acc: 0.0551 - val_loss: 4.5419 - val_acc: 0.0419
Epoch 19/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.4280 - acc: 0.0577Epoch 00019: val_loss improved from 4.54191 to 4.53347, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 21s 3ms/step - loss: 4.4283 - acc: 0.0575 - val_loss: 4.5335 - val_acc: 0.0467
Epoch 20/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.4057 - acc: 0.0574Epoch 00020: val_loss did not improve
6680/6680 [==============================] - 21s 3ms/step - loss: 4.4066 - acc: 0.0572 - val_loss: 4.5596 - val_acc: 0.0443
Out[15]:
<keras.callbacks.History at 0x7f21cef948d0>

Load the Model with the Best Validation Loss


In [16]:
model.load_weights('saved_models/weights.best.from_scratch.hdf5')

Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 1%.


In [17]:
# get index of predicted dog breed for each image in test set
dog_breed_predictions = [np.argmax(model.predict(np.expand_dims(tensor, axis=0))) for tensor in test_tensors]

# report test accuracy
test_accuracy = 100*np.sum(np.array(dog_breed_predictions)==np.argmax(test_targets, axis=1))/len(dog_breed_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)


Test accuracy: 4.9043%

Step 4: Use a CNN to Classify Dog Breeds

To reduce training time without sacrificing accuracy, we show you how to train a CNN using transfer learning. In the following step, you will get a chance to use transfer learning to train your own CNN.

Obtain Bottleneck Features


In [18]:
bottleneck_features = np.load('/data/bottleneck_features/DogVGG16Data.npz')
train_VGG16 = bottleneck_features['train']
valid_VGG16 = bottleneck_features['valid']
test_VGG16 = bottleneck_features['test']

Model Architecture

The model uses the the pre-trained VGG-16 model as a fixed feature extractor, where the last convolutional output of VGG-16 is fed as input to our model. We only add a global average pooling layer and a fully connected layer, where the latter contains one node for each dog category and is equipped with a softmax.


In [19]:
VGG16_model = Sequential()
VGG16_model.add(GlobalAveragePooling2D(input_shape=train_VGG16.shape[1:]))
VGG16_model.add(Dense(133, activation='softmax'))

VGG16_model.summary()


_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_2 ( (None, 512)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 133)               68229     
=================================================================
Total params: 68,229
Trainable params: 68,229
Non-trainable params: 0
_________________________________________________________________

Compile the Model


In [20]:
VGG16_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])

Train the Model


In [21]:
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.VGG16.hdf5', 
                               verbose=1, save_best_only=True)

VGG16_model.fit(train_VGG16, train_targets, 
          validation_data=(valid_VGG16, valid_targets),
          epochs=20, batch_size=20, callbacks=[checkpointer], verbose=1)


Train on 6680 samples, validate on 835 samples
Epoch 1/20
6460/6680 [============================>.] - ETA: 0s - loss: 11.9395 - acc: 0.1331Epoch 00001: val_loss improved from inf to 10.31732, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 302us/step - loss: 11.8793 - acc: 0.1370 - val_loss: 10.3173 - val_acc: 0.2371
Epoch 2/20
6640/6680 [============================>.] - ETA: 0s - loss: 9.5599 - acc: 0.3102Epoch 00002: val_loss improved from 10.31732 to 9.46880, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 256us/step - loss: 9.5466 - acc: 0.3109 - val_loss: 9.4688 - val_acc: 0.3150
Epoch 3/20
6440/6680 [===========================>..] - ETA: 0s - loss: 9.0361 - acc: 0.3793Epoch 00003: val_loss improved from 9.46880 to 9.26609, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 248us/step - loss: 9.0164 - acc: 0.3813 - val_loss: 9.2661 - val_acc: 0.3533
Epoch 4/20
6440/6680 [===========================>..] - ETA: 0s - loss: 8.7822 - acc: 0.4101Epoch 00004: val_loss improved from 9.26609 to 9.15203, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 244us/step - loss: 8.8160 - acc: 0.4084 - val_loss: 9.1520 - val_acc: 0.3617
Epoch 5/20
6580/6680 [============================>.] - ETA: 0s - loss: 8.6295 - acc: 0.4281Epoch 00005: val_loss improved from 9.15203 to 9.04905, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 246us/step - loss: 8.6328 - acc: 0.4278 - val_loss: 9.0490 - val_acc: 0.3808
Epoch 6/20
6620/6680 [============================>.] - ETA: 0s - loss: 8.4837 - acc: 0.4429Epoch 00006: val_loss improved from 9.04905 to 8.87806, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 245us/step - loss: 8.4809 - acc: 0.4428 - val_loss: 8.8781 - val_acc: 0.3772
Epoch 7/20
6640/6680 [============================>.] - ETA: 0s - loss: 8.2715 - acc: 0.4559Epoch 00007: val_loss improved from 8.87806 to 8.75192, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 244us/step - loss: 8.2755 - acc: 0.4555 - val_loss: 8.7519 - val_acc: 0.3988
Epoch 8/20
6640/6680 [============================>.] - ETA: 0s - loss: 8.1337 - acc: 0.4694Epoch 00008: val_loss improved from 8.75192 to 8.68967, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 244us/step - loss: 8.1262 - acc: 0.4699 - val_loss: 8.6897 - val_acc: 0.3964
Epoch 9/20
6580/6680 [============================>.] - ETA: 0s - loss: 7.9846 - acc: 0.4827Epoch 00009: val_loss improved from 8.68967 to 8.47862, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 247us/step - loss: 7.9737 - acc: 0.4834 - val_loss: 8.4786 - val_acc: 0.4156
Epoch 10/20
6600/6680 [============================>.] - ETA: 0s - loss: 7.8437 - acc: 0.4958Epoch 00010: val_loss improved from 8.47862 to 8.33525, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 245us/step - loss: 7.8541 - acc: 0.4952 - val_loss: 8.3353 - val_acc: 0.4204
Epoch 11/20
6660/6680 [============================>.] - ETA: 0s - loss: 7.7343 - acc: 0.5063Epoch 00011: val_loss did not improve
6680/6680 [==============================] - 2s 243us/step - loss: 7.7377 - acc: 0.5061 - val_loss: 8.4024 - val_acc: 0.4168
Epoch 12/20
6460/6680 [============================>.] - ETA: 0s - loss: 7.6952 - acc: 0.5115Epoch 00012: val_loss did not improve
6680/6680 [==============================] - 2s 241us/step - loss: 7.7170 - acc: 0.5099 - val_loss: 8.3620 - val_acc: 0.4132
Epoch 13/20
6460/6680 [============================>.] - ETA: 0s - loss: 7.6580 - acc: 0.5169Epoch 00013: val_loss improved from 8.33525 to 8.29859, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 243us/step - loss: 7.6659 - acc: 0.5162 - val_loss: 8.2986 - val_acc: 0.4263
Epoch 14/20
6580/6680 [============================>.] - ETA: 0s - loss: 7.5778 - acc: 0.5220Epoch 00014: val_loss improved from 8.29859 to 8.19734, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 245us/step - loss: 7.5752 - acc: 0.5219 - val_loss: 8.1973 - val_acc: 0.4287
Epoch 15/20
6560/6680 [============================>.] - ETA: 0s - loss: 7.2865 - acc: 0.5290Epoch 00015: val_loss improved from 8.19734 to 7.90474, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 244us/step - loss: 7.2680 - acc: 0.5298 - val_loss: 7.9047 - val_acc: 0.4371
Epoch 16/20
6640/6680 [============================>.] - ETA: 0s - loss: 7.1073 - acc: 0.5465Epoch 00016: val_loss improved from 7.90474 to 7.77973, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 244us/step - loss: 7.1038 - acc: 0.5467 - val_loss: 7.7797 - val_acc: 0.4599
Epoch 17/20
6600/6680 [============================>.] - ETA: 0s - loss: 7.0684 - acc: 0.5508Epoch 00017: val_loss improved from 7.77973 to 7.77492, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 244us/step - loss: 7.0542 - acc: 0.5516 - val_loss: 7.7749 - val_acc: 0.4599
Epoch 18/20
6520/6680 [============================>.] - ETA: 0s - loss: 6.9776 - acc: 0.5572Epoch 00018: val_loss did not improve
6680/6680 [==============================] - 2s 246us/step - loss: 6.9743 - acc: 0.5573 - val_loss: 7.8058 - val_acc: 0.4551
Epoch 19/20
6620/6680 [============================>.] - ETA: 0s - loss: 6.9206 - acc: 0.5622Epoch 00019: val_loss improved from 7.77492 to 7.71022, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 245us/step - loss: 6.9407 - acc: 0.5609 - val_loss: 7.7102 - val_acc: 0.4659
Epoch 20/20
6640/6680 [============================>.] - ETA: 0s - loss: 6.8868 - acc: 0.5642Epoch 00020: val_loss did not improve
6680/6680 [==============================] - 2s 244us/step - loss: 6.8890 - acc: 0.5641 - val_loss: 7.7134 - val_acc: 0.4671
Out[21]:
<keras.callbacks.History at 0x7f21cee42a58>

Load the Model with the Best Validation Loss


In [22]:
VGG16_model.load_weights('saved_models/weights.best.VGG16.hdf5')

Test the Model

Now, we can use the CNN to test how well it identifies breed within our test dataset of dog images. We print the test accuracy below.


In [23]:
# get index of predicted dog breed for each image in test set
VGG16_predictions = [np.argmax(VGG16_model.predict(np.expand_dims(feature, axis=0))) for feature in test_VGG16]

# report test accuracy
test_accuracy = 100*np.sum(np.array(VGG16_predictions)==np.argmax(test_targets, axis=1))/len(VGG16_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)


Test accuracy: 45.0957%

Predict Dog Breed with the Model


In [24]:
from extract_bottleneck_features import *

def VGG16_predict_breed(img_path):
    # extract bottleneck features
    bottleneck_feature = extract_VGG16(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = VGG16_model.predict(bottleneck_feature)
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)

You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.

In Step 4, we used transfer learning to create a CNN using VGG-16 bottleneck features. In this section, you must use the bottleneck features from a different pre-trained model. To make things easier for you, we have pre-computed the features for all of the networks that are currently available in Keras. These are already in the workspace, at /data/bottleneck_features. If you wish to download them on a different machine, they can be found at:

The files are encoded as such:

Dog{network}Data.npz

where {network}, in the above filename, can be one of VGG19, Resnet50, InceptionV3, or Xception.

The above architectures are downloaded and stored for you in the /data/bottleneck_features/ folder.

This means the following will be in the /data/bottleneck_features/ folder:

DogVGG19Data.npz DogResnet50Data.npz DogInceptionV3Data.npz DogXceptionData.npz

(IMPLEMENTATION) Obtain Bottleneck Features

In the code block below, extract the bottleneck features corresponding to the train, test, and validation sets by running the following:

bottleneck_features = np.load('/data/bottleneck_features/Dog{network}Data.npz')
train_{network} = bottleneck_features['train']
valid_{network} = bottleneck_features['valid']
test_{network} = bottleneck_features['test']

In [68]:
### TODO: Obtain bottleneck features from another pre-trained CNN.
bottleneck_features = np.load('/data/bottleneck_features/DogXceptionData.npz')
train_Xception = bottleneck_features['train']
valid_Xception = bottleneck_features['valid']
test_Xception = bottleneck_features['test']

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    <your model's name>.summary()

Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.

Answer: The proposed CNN architecture uses cross learned features from V19 serialized model provided to us. I use a softmax function at the end to be able to specify the loss type and predict the coutcome based on activation from previous layers.

Training the model and developing a well performing deep architecture is a time consuming task in deep learning. The cost is significant both in terms of developer time and CPU time. As the model becomes complex, cost of training grows non-linearly. Complex models need to be developed when simple models do not provide high accuracy. A complex deep learning model can be developed only if one has lot of experience in neural network architectures. Transfer learning approach is advantageous, as it allows for the lower layers to be shared between the developers and systems. The use of this technique allows us to over come the problem of limited labeled data in a domain compared to another domain. Helps us to iterate faster in development of deep networks.


In [69]:
### TODO: Define your architecture.
Xception_model = Sequential()
Xception_model.add(GlobalAveragePooling2D(input_shape=train_VGG19.shape[1:]))
Xception_model.add(Dense(133, activation='softmax'))
Xception_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', 
                  metrics=['accuracy'])
Xception_model.summary()


_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_8 ( (None, 2048)              0         
_________________________________________________________________
dense_10 (Dense)             (None, 133)               272517    
=================================================================
Total params: 272,517
Trainable params: 272,517
Non-trainable params: 0
_________________________________________________________________

(IMPLEMENTATION) Compile the Model


In [71]:
### TODO: Compile the model.
Xception_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.


In [72]:
from keras.callbacks import ModelCheckpoint  
### TODO: Train the model.
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.Xception.hdf5', 
                               verbose=1, save_best_only=True)

history = Xception_model.fit(train_Xception, train_targets, 
          validation_data=(valid_Xception, valid_targets),
          epochs=50, batch_size=100, callbacks=[checkpointer], verbose=1)


Train on 6680 samples, validate on 835 samples
Epoch 1/50
6500/6680 [============================>.] - ETA: 0s - loss: 1.5958 - acc: 0.6855Epoch 00001: val_loss improved from inf to 0.67491, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 4s 560us/step - loss: 1.5728 - acc: 0.6883 - val_loss: 0.6749 - val_acc: 0.8204
Epoch 2/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.4882 - acc: 0.8689Epoch 00002: val_loss improved from 0.67491 to 0.51321, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 3s 451us/step - loss: 0.4850 - acc: 0.8698 - val_loss: 0.5132 - val_acc: 0.8395
Epoch 3/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.3376 - acc: 0.9026Epoch 00003: val_loss improved from 0.51321 to 0.47311, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 3s 463us/step - loss: 0.3396 - acc: 0.9016 - val_loss: 0.4731 - val_acc: 0.8371
Epoch 4/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.2655 - acc: 0.9208Epoch 00004: val_loss improved from 0.47311 to 0.45920, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 3s 476us/step - loss: 0.2694 - acc: 0.9202 - val_loss: 0.4592 - val_acc: 0.8467
Epoch 5/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.2155 - acc: 0.9362Epoch 00005: val_loss improved from 0.45920 to 0.45649, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 3s 481us/step - loss: 0.2171 - acc: 0.9353 - val_loss: 0.4565 - val_acc: 0.8503
Epoch 6/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.1804 - acc: 0.9486Epoch 00006: val_loss improved from 0.45649 to 0.44882, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 3s 483us/step - loss: 0.1795 - acc: 0.9491 - val_loss: 0.4488 - val_acc: 0.8527
Epoch 7/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.1511 - acc: 0.9591Epoch 00007: val_loss did not improve
6680/6680 [==============================] - 3s 478us/step - loss: 0.1516 - acc: 0.9584 - val_loss: 0.4520 - val_acc: 0.8587
Epoch 8/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.1285 - acc: 0.9660Epoch 00008: val_loss did not improve
6680/6680 [==============================] - 3s 464us/step - loss: 0.1278 - acc: 0.9662 - val_loss: 0.4547 - val_acc: 0.8563
Epoch 9/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.1095 - acc: 0.9723Epoch 00009: val_loss did not improve
6680/6680 [==============================] - 3s 457us/step - loss: 0.1100 - acc: 0.9722 - val_loss: 0.4604 - val_acc: 0.8623
Epoch 10/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0923 - acc: 0.9780Epoch 00010: val_loss did not improve
6680/6680 [==============================] - 3s 473us/step - loss: 0.0930 - acc: 0.9778 - val_loss: 0.4649 - val_acc: 0.8611
Epoch 11/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0802 - acc: 0.9820Epoch 00011: val_loss did not improve
6680/6680 [==============================] - 3s 479us/step - loss: 0.0799 - acc: 0.9819 - val_loss: 0.4558 - val_acc: 0.8623
Epoch 12/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0691 - acc: 0.9855Epoch 00012: val_loss did not improve
6680/6680 [==============================] - 3s 472us/step - loss: 0.0688 - acc: 0.9855 - val_loss: 0.4711 - val_acc: 0.8587
Epoch 13/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0593 - acc: 0.9889Epoch 00013: val_loss did not improve
6680/6680 [==============================] - 3s 460us/step - loss: 0.0594 - acc: 0.9885 - val_loss: 0.4845 - val_acc: 0.8587
Epoch 14/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0537 - acc: 0.9885Epoch 00014: val_loss did not improve
6680/6680 [==============================] - 3s 460us/step - loss: 0.0534 - acc: 0.9885 - val_loss: 0.4895 - val_acc: 0.8503
Epoch 15/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0452 - acc: 0.9906Epoch 00015: val_loss did not improve
6680/6680 [==============================] - 3s 475us/step - loss: 0.0460 - acc: 0.9903 - val_loss: 0.5011 - val_acc: 0.8503
Epoch 16/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0408 - acc: 0.9925Epoch 00016: val_loss did not improve
6680/6680 [==============================] - 3s 480us/step - loss: 0.0413 - acc: 0.9922 - val_loss: 0.5104 - val_acc: 0.8587
Epoch 17/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0356 - acc: 0.9929Epoch 00017: val_loss did not improve
6680/6680 [==============================] - 3s 487us/step - loss: 0.0363 - acc: 0.9928 - val_loss: 0.5136 - val_acc: 0.8635
Epoch 18/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0321 - acc: 0.9940Epoch 00018: val_loss did not improve
6680/6680 [==============================] - 3s 482us/step - loss: 0.0324 - acc: 0.9937 - val_loss: 0.5229 - val_acc: 0.8563
Epoch 19/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0291 - acc: 0.9952Epoch 00019: val_loss did not improve
6680/6680 [==============================] - 3s 482us/step - loss: 0.0292 - acc: 0.9951 - val_loss: 0.5231 - val_acc: 0.8611
Epoch 20/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0258 - acc: 0.9955Epoch 00020: val_loss did not improve
6680/6680 [==============================] - 3s 486us/step - loss: 0.0257 - acc: 0.9957 - val_loss: 0.5288 - val_acc: 0.8611
Epoch 21/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0225 - acc: 0.9972Epoch 00021: val_loss did not improve
6680/6680 [==============================] - 3s 484us/step - loss: 0.0229 - acc: 0.9970 - val_loss: 0.5356 - val_acc: 0.8623
Epoch 22/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0205 - acc: 0.9968Epoch 00022: val_loss did not improve
6680/6680 [==============================] - 3s 468us/step - loss: 0.0206 - acc: 0.9966 - val_loss: 0.5526 - val_acc: 0.8599
Epoch 23/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0197 - acc: 0.9969Epoch 00023: val_loss did not improve
6680/6680 [==============================] - 3s 482us/step - loss: 0.0196 - acc: 0.9970 - val_loss: 0.5671 - val_acc: 0.8611
Epoch 24/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0172 - acc: 0.9968Epoch 00024: val_loss did not improve
6680/6680 [==============================] - 3s 480us/step - loss: 0.0172 - acc: 0.9967 - val_loss: 0.5592 - val_acc: 0.8599
Epoch 25/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0160 - acc: 0.9971Epoch 00025: val_loss did not improve
6680/6680 [==============================] - 3s 479us/step - loss: 0.0158 - acc: 0.9972 - val_loss: 0.5645 - val_acc: 0.8539
Epoch 26/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0146 - acc: 0.9975Epoch 00026: val_loss did not improve
6680/6680 [==============================] - 3s 460us/step - loss: 0.0145 - acc: 0.9976 - val_loss: 0.5747 - val_acc: 0.8599
Epoch 27/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0135 - acc: 0.9980Epoch 00027: val_loss did not improve
6680/6680 [==============================] - 3s 448us/step - loss: 0.0132 - acc: 0.9981 - val_loss: 0.5857 - val_acc: 0.8623
Epoch 28/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0125 - acc: 0.9980Epoch 00028: val_loss did not improve
6680/6680 [==============================] - 3s 456us/step - loss: 0.0123 - acc: 0.9981 - val_loss: 0.6011 - val_acc: 0.8551
Epoch 29/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0102 - acc: 0.9978Epoch 00029: val_loss did not improve
6680/6680 [==============================] - 3s 469us/step - loss: 0.0109 - acc: 0.9978 - val_loss: 0.6007 - val_acc: 0.8611
Epoch 30/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0103 - acc: 0.9983Epoch 00030: val_loss did not improve
6680/6680 [==============================] - 3s 457us/step - loss: 0.0103 - acc: 0.9982 - val_loss: 0.6104 - val_acc: 0.8527
Epoch 31/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0096 - acc: 0.9983Epoch 00031: val_loss did not improve
6680/6680 [==============================] - 3s 457us/step - loss: 0.0094 - acc: 0.9984 - val_loss: 0.6204 - val_acc: 0.8587
Epoch 32/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0096 - acc: 0.9982Epoch 00032: val_loss did not improve
6680/6680 [==============================] - 3s 459us/step - loss: 0.0095 - acc: 0.9982 - val_loss: 0.6286 - val_acc: 0.8623
Epoch 33/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0082 - acc: 0.9985Epoch 00033: val_loss did not improve
6680/6680 [==============================] - 3s 469us/step - loss: 0.0081 - acc: 0.9985 - val_loss: 0.6352 - val_acc: 0.8587
Epoch 34/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0077 - acc: 0.9983Epoch 00034: val_loss did not improve
6680/6680 [==============================] - 3s 458us/step - loss: 0.0076 - acc: 0.9984 - val_loss: 0.6337 - val_acc: 0.8599
Epoch 35/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0068 - acc: 0.9986Epoch 00035: val_loss did not improve
6680/6680 [==============================] - 3s 451us/step - loss: 0.0068 - acc: 0.9987 - val_loss: 0.6737 - val_acc: 0.8563
Epoch 36/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0069 - acc: 0.9988Epoch 00036: val_loss did not improve
6680/6680 [==============================] - 3s 455us/step - loss: 0.0070 - acc: 0.9987 - val_loss: 0.6564 - val_acc: 0.8659
Epoch 37/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0062 - acc: 0.9985Epoch 00037: val_loss did not improve
6680/6680 [==============================] - 3s 480us/step - loss: 0.0063 - acc: 0.9985 - val_loss: 0.6635 - val_acc: 0.8515
Epoch 38/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0052 - acc: 0.9988Epoch 00038: val_loss did not improve
6680/6680 [==============================] - 3s 483us/step - loss: 0.0060 - acc: 0.9985 - val_loss: 0.6907 - val_acc: 0.8575
Epoch 39/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0053 - acc: 0.9986Epoch 00039: val_loss did not improve
6680/6680 [==============================] - 3s 482us/step - loss: 0.0053 - acc: 0.9987 - val_loss: 0.6694 - val_acc: 0.8587
Epoch 40/50
6600/6680 [============================>.] - ETA: 0s - loss: 0.0055 - acc: 0.9986Epoch 00040: val_loss did not improve
6680/6680 [==============================] - 3s 489us/step - loss: 0.0055 - acc: 0.9985 - val_loss: 0.7018 - val_acc: 0.8539
Epoch 41/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0051 - acc: 0.9985Epoch 00041: val_loss did not improve
6680/6680 [==============================] - 3s 480us/step - loss: 0.0050 - acc: 0.9985 - val_loss: 0.7037 - val_acc: 0.8587
Epoch 42/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0048 - acc: 0.9985Epoch 00042: val_loss did not improve
6680/6680 [==============================] - 3s 483us/step - loss: 0.0048 - acc: 0.9985 - val_loss: 0.7243 - val_acc: 0.8527
Epoch 43/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0049 - acc: 0.9983Epoch 00043: val_loss did not improve
6680/6680 [==============================] - 3s 478us/step - loss: 0.0048 - acc: 0.9984 - val_loss: 0.7220 - val_acc: 0.8587
Epoch 44/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0045 - acc: 0.9985Epoch 00044: val_loss did not improve
6680/6680 [==============================] - 3s 502us/step - loss: 0.0045 - acc: 0.9985 - val_loss: 0.7266 - val_acc: 0.8539
Epoch 45/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0042 - acc: 0.9988Epoch 00045: val_loss did not improve
6680/6680 [==============================] - 3s 494us/step - loss: 0.0041 - acc: 0.9988 - val_loss: 0.7533 - val_acc: 0.8551
Epoch 46/50
6600/6680 [============================>.] - ETA: 0s - loss: 0.0042 - acc: 0.9986Epoch 00046: val_loss did not improve
6680/6680 [==============================] - 3s 499us/step - loss: 0.0042 - acc: 0.9987 - val_loss: 0.7409 - val_acc: 0.8539
Epoch 47/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0033 - acc: 0.9985Epoch 00047: val_loss did not improve
6680/6680 [==============================] - 3s 498us/step - loss: 0.0038 - acc: 0.9984 - val_loss: 0.7512 - val_acc: 0.8539
Epoch 48/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0037 - acc: 0.9983Epoch 00048: val_loss did not improve
6680/6680 [==============================] - 3s 485us/step - loss: 0.0038 - acc: 0.9982 - val_loss: 0.7531 - val_acc: 0.8551
Epoch 49/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0034 - acc: 0.9988Epoch 00049: val_loss did not improve
6680/6680 [==============================] - 3s 467us/step - loss: 0.0034 - acc: 0.9988 - val_loss: 0.7606 - val_acc: 0.8527
Epoch 50/50
6500/6680 [============================>.] - ETA: 0s - loss: 0.0037 - acc: 0.9988Epoch 00050: val_loss did not improve
6680/6680 [==============================] - 3s 478us/step - loss: 0.0036 - acc: 0.9988 - val_loss: 0.7793 - val_acc: 0.8551

(IMPLEMENTATION) Load the Model with the Best Validation Loss


In [73]:
### TODO: Load the model weights with the best validation loss.
Xception_model.load_weights('saved_models/weights.best.Xception.hdf5')

(IMPLEMENTATION) Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 60%.


In [74]:
### TODO: Calculate classification accuracy on the test dataset.
# get index of predicted dog breed for each image in test set
Xception_predictions = [np.argmax(Xception_model.predict(np.expand_dims(feature, axis=0))) for feature in test_Xception]

# report test accuracy
test_accuracy = 100*np.sum(np.array(Xception_predictions)==np.argmax(test_targets, axis=1))/len(Xception_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)


Test accuracy: 85.0478%

(IMPLEMENTATION) Predict Dog Breed with the Model

Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan_hound, etc) that is predicted by your model.

Similar to the analogous function in Step 5, your function should have three steps:

  1. Extract the bottleneck features corresponding to the chosen CNN model.
  2. Supply the bottleneck features as input to the model to return the predicted vector. Note that the argmax of this prediction vector gives the index of the predicted dog breed.
  3. Use the dog_names array defined in Step 0 of this notebook to return the corresponding breed.

The functions to extract the bottleneck features can be found in extract_bottleneck_features.py, and they have been imported in an earlier code cell. To obtain the bottleneck features corresponding to your chosen CNN architecture, you need to use the function

extract_{network}

where {network}, in the above filename, should be one of VGG19, Resnet50, InceptionV3, or Xception.


In [75]:
### TODO: Write a function that takes a path to an image as input
### and returns the dog breed that is predicted by the model.
def Xception_predict_breed(img_path):
    # extract bottleneck features
    bottleneck_feature = extract_Xception(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = Xception_model.predict(bottleneck_feature)
    # return dog breed that is predicted by the model
    return (dog_names[np.argmax(predicted_vector)], predicted_vector[0][np.argmax(predicted_vector)])

In [76]:
import glob
from PIL import Image
from io import BytesIO
from IPython.display import HTML

def get_thumbnail(path):
    i = Image.open(path).convert('RGB')
    i.thumbnail((150, 150), Image.LANCZOS)
    return i

def image_base64(im):
    if isinstance(im, str):
        im = get_thumbnail(im)
    with BytesIO() as buffer:
        im.save(buffer, 'jpeg')
        return base64.b64encode(buffer.getvalue()).decode()

def image_formatter(im):
    return f'<img src="data:image/jpeg;base64,{image_base64(im)}">'

def file_formatter(file_name):
    return f'<img src="{file_name}" width="150" height="150">'

predictions_and_images = dict()
for filename in glob.iglob('dogs/*'):
    label, score = Xception_predict_breed(filename)
    print("Predicting that image in  {0}, depicts a {1}, with a score {2}".format(filename, label, score))
    predictions_and_images[filename] = label


Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.4/xception_weights_tf_dim_ordering_tf_kernels_notop.h5
83689472/83683744 [==============================] - 1s 0us/step
Predicting that image in  dogs/GermanSheperd.jpg, depicts a in/071.German_shepherd_dog, with a score 0.9938569664955139
Predicting that image in  dogs/doberman.jpg, depicts a in/059.Doberman_pinscher, with a score 0.5949955582618713
Predicting that image in  dogs/afghan-hound.jpg, depicts a in/002.Afghan_hound, with a score 0.9960762858390808
Predicting that image in  dogs/maltese.jpg, depicts a in/082.Havanese, with a score 0.4844890534877777
Predicting that image in  dogs/great-dane.jpg, depicts a in/081.Greyhound, with a score 0.9456342458724976
Predicting that image in  dogs/chiwawa.jpg, depicts a in/048.Chihuahua, with a score 0.8339229822158813
Predicting that image in  dogs/Labrador-Retriever-On-White-07.jpg, depicts a in/096.Labrador_retriever, with a score 0.5718789100646973

In [77]:
import pandas as pd
import base64
from IPython.display import HTML
predictions_and_images
predictions_df = pd.DataFrame.from_dict(predictions_and_images, orient='index')
predictions_df.reset_index(inplace=True)
predictions_df.columns = ['file', 'label']
predictions_df['image'] = predictions_df.file.map(lambda f: get_thumbnail(f))
predictions_df
HTML(predictions_df[['label', 'file']].to_html(formatters={'file': file_formatter}, escape=False))


Out[77]:
label file
0 in/071.German_shepherd_dog
1 in/059.Doberman_pinscher
2 in/002.Afghan_hound
3 in/082.Havanese
4 in/081.Greyhound
5 in/048.Chihuahua
6 in/096.Labrador_retriever

Step 6: Write your Algorithm

Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

  • if a dog is detected in the image, return the predicted breed.
  • if a human is detected in the image, return the resembling dog breed.
  • if neither is detected in the image, provide output that indicates an error.

You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and dog_detector functions developed above. You are required to use your CNN from Step 5 to predict dog breed.

Some sample output for our algorithm is provided below, but feel free to design your own user experience!

(IMPLEMENTATION) Write your Algorithm


In [78]:
### TODO: Write your algorithm.
### Feel free to use as many code cells as needed.
    
def predict_for_image(filename):
    label, score = Xception_predict_breed(filename)
    if (score > 0.95) :
        predictions_and_images[filename] = ('dog', label, score)
    elif face_detector(filename):
        predictions_and_images[filename] = ('human', label, score)
    else:
        predictions_and_images[filename] = ('neither', label, score)
    return predictions_and_images

Step 7: Test Your Algorithm

In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?

(IMPLEMENTATION) Test Your Algorithm on Sample Images!

Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.

Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.

Answer: The algorithm is worst then I expected. For each of the predictions, we need a confidence score associated with the prediction. As it currently stands the label does not capture contents of the image and all the images with humans/dogs are allocated a breed. This problem may be symptomatic of the way the label space is created. The three ways that I would fix the problem are:

  • Have a 'None' label to denote that the image does not contain any dog breeds
  • Include confidence intervals with each of the predictions
  • Improve on classifier accuracy

In [79]:
## Execute your algorithm from Step 6 on
## at least 6 images on your computer.
## Feel free to use as many code cells as needed.
for filename in glob.iglob('random_images/*'):
    type_of_animal, label, score = predict_for_image(filename)[filename]
    print("Predicting that image in  {0}, depicts a {1}, closest dog breed {2}, score = {3}".format(filename, type_of_animal, label, score))


Predicting that image in  random_images/golden-retreiver.jpg, depicts a dog, closest dog breed in/076.Golden_retriever, score = 0.999826967716217
Predicting that image in  random_images/girl.jpg, depicts a human, closest dog breed in/056.Dachshund, score = 0.10800057649612427
Predicting that image in  random_images/woman.jpg, depicts a human, closest dog breed in/056.Dachshund, score = 0.12749311327934265
Predicting that image in  random_images/house-n-tree.jpg, depicts a neither, closest dog breed in/100.Lowchen, score = 0.06626333296298981
Predicting that image in  random_images/River.jpg, depicts a neither, closest dog breed in/010.Anatolian_shepherd_dog, score = 0.08305960893630981
Predicting that image in  random_images/Doberman-and-human.jpg, depicts a human, closest dog breed in/059.Doberman_pinscher, score = 0.5463489294052124

Please download your notebook to submit


In [ ]: