# Artificial Intelligence Nanodegree¶

## 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.

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!

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
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('dogImages/train')
valid_files, valid_targets = load_dataset('dogImages/valid')
test_files, test_targets = load_dataset('dogImages/test')

# load list of dog names
dog_names = [item[20:-1] for item in sorted(glob("dogImages/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("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

# load color (BGR) image
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# find faces in image

# 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):
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
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.

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.
count = 0
for img_path in human_files_short:
if face_detector(img_path):
count += 1
human_face_percentage = 100*count/len(human_files_short)
print('Faces detected in human files: {}%'.format(human_face_percentage))

count = 0
for img_path in dog_files_short:
if face_detector(img_path):
count += 1
dog_face_percentage = 100*count/len(dog_files_short)

print('Faces detected in dog files: {}%'.format(dog_face_percentage))

Faces detected in human files: 99.0%
Faces detected in dog files: 11.0%

In [6]:
img = cv2.imread(human_files[0])
print(human_files_short[0])
print(face_detector(human_files_short[0]))
# 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()

lfw\Muhammad_Ali\Muhammad_Ali_0001.jpg
False


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: It is somewhat reasonable that users supply a direct figure of their face since the point is to determine what type of dog they look like. It would be hard to determine that if we can't see their face. However, if you look at the first image of the human data set, Mohammad Ali's face is visible, but it is at an angle. The face detector doesn't detect it as a face, but the face is visible enough that you could make a guess at what type of dog he looks like. The app should be able to handle that type of image.

Looking at the dog data, we see that there are face detections in 11% of the data. Looking through some of the dog data I was able to find a few pictures where there were humans with the dogs. If the app just uses a face detector to check if it is an image of a human then it is going to say that any image with a human in it is an image of a human.

Based on all of this if our goal is to make a human detector I would say a better approach would be to just use a pretrained network to detect if a human is in the image. We could use the dog and human data and use a 2 node output layer corresponding to human or dog, but this would run into problems if the image supplied did not have either. I think the best option would be to train two separate networks. One to detect humans and one to detect dogs. To do this though you would need some data without dogs and without humans. This, might me something to try out in the future, but for now I think the face detector does a pretty good job detecting human faces and the user will just have to be instructed to take quality images.

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.

In [7]:
## (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')


### 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 [168]:
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 [169]:
### 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?

In [170]:
### TODO: Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.
count = 0
index = 0
for img_path in human_files_short:
if dog_detector(img_path):
count += 1
print('Dog detected at index', index)
print(img_path)
index += 1
human_dog_percentage = 100*count/len(human_files_short)
print('Dogs detected in human files: {}%'.format(human_dog_percentage))

count = 0
for img_path in dog_files_short:
if dog_detector(img_path):
count += 1
dog_dog_percentage = 100*count/len(dog_files_short)

print('Dogs detected in dog files: {}%'.format(dog_dog_percentage))

Dog detected at index 44
lfw\John_Rigas\John_Rigas_0001.jpg
Dogs detected in human files: 1.0%
Dogs detected in dog files: 100.0%

In [171]:
img = cv2.imread(human_files[44])
print(human_files_short[44])
print(dog_detector(human_files_short[44]))
prediction = ResNet50_predict_labels(human_files_short[44])
print(prediction)
# 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()

lfw\John_Rigas\John_Rigas_0001.jpg
True
190


Apparently the dog detector thinks John Rigas is a dog. Looking at the prediction from the ResNet model, it's classifying him as a Sealyham terrier. Not really seeing it, but the dog detector only has 1% error so it's not doing too bad.

## 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.

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 [13]:
train_files[:10]

Out[13]:
array(['dogImages/train\\095.Kuvasz\\Kuvasz_06442.jpg',
'dogImages/train\\057.Dalmatian\\Dalmatian_04054.jpg',
'dogImages/train\\088.Irish_water_spaniel\\Irish_water_spaniel_06014.jpg',
'dogImages/train\\008.American_staffordshire_terrier\\American_staffordshire_terrier_00596.jpg',
'dogImages/train\\008.American_staffordshire_terrier\\American_staffordshire_terrier_00563.jpg',
'dogImages/train\\063.English_springer_spaniel\\English_springer_spaniel_04469.jpg',
'dogImages/train\\054.Collie\\Collie_03789.jpg',
'dogImages/train\\119.Petit_basset_griffon_vendeen\\Petit_basset_griffon_vendeen_07705.jpg',
'dogImages/train\\009.American_water_spaniel\\American_water_spaniel_00625.jpg',
'dogImages/train\\081.Greyhound\\Greyhound_05559.jpg'],
dtype='<U99')
In [14]:
from PIL import ImageFile

In [15]:
# 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 [00:54<00:00, 123.53it/s]
100%|███████████████████████████████████████████████████████████████████████████████| 835/835 [00:06<00:00, 125.09it/s]
100%|███████████████████████████████████████████████████████████████████████████████| 836/836 [00:06<00:00, 127.29it/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 chose to use an architecture similar to the VGG16 network, only using one convolutional lay for each convolutional block. I chose to use the suggested GAP and single output layer, but with a BatchNormalization layer to normalize the output from the GAP layer. The single output layer performed about as good as when adding other dense layers, but took less time to train and was less prone to overfitting.

In [20]:
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from keras.layers import Dropout, Flatten, Dense, BatchNormalization
from keras.models import Sequential

### TODO: Define your architecture.
model = Sequential()
# Conv layer1

# Conv layer2

# Conv layer3

# Conv layer4

# Conv layer5

#Flatten Layer

#Fully Connected Layer 2

model.summary()

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
conv2d_1 (Conv2D)            (None, 224, 224, 32)      896
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 112, 112, 32)      0
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 112, 112, 64)      18496
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 56, 56, 64)        0
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 56, 56, 128)       73856
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 28, 28, 128)       0
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 28, 28, 256)       295168
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 14, 14, 256)       0
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 14, 14, 256)       590080
_________________________________________________________________
max_pooling2d_6 (MaxPooling2 (None, 7, 7, 256)         0
_________________________________________________________________
global_average_pooling2d_1 ( (None, 256)               0
_________________________________________________________________
batch_normalization_1 (Batch (None, 256)               1024
_________________________________________________________________
dense_1 (Dense)              (None, 133)               34181
=================================================================
Total params: 1,013,701
Trainable params: 1,013,189
Non-trainable params: 512
_________________________________________________________________

In [87]:
model.save_weights('saved_models/weights.initial_scratch_model.hdf5')


### Compile the Model¶

In [27]:
from keras.optimizers import Adam


### (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 [61]:
from keras.callbacks import ModelCheckpoint

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

epochs = 10

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

checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch_no_da.hdf5',
verbose=0, save_best_only=True)

model.fit(train_tensors, train_targets, batch_size=32, epochs=epochs,
callbacks=[checkpointer], verbose=2,
validation_data=(valid_tensors, valid_targets))

Train on 6680 samples, validate on 835 samples
Epoch 1/10
- 25s - loss: 4.7900 - acc: 0.0157 - val_loss: 4.8234 - val_acc: 0.0144
Epoch 2/10
- 23s - loss: 4.5528 - acc: 0.0334 - val_loss: 4.7471 - val_acc: 0.0263
Epoch 3/10
- 23s - loss: 4.3038 - acc: 0.0581 - val_loss: 5.0657 - val_acc: 0.0263
Epoch 4/10
- 23s - loss: 4.0407 - acc: 0.0870 - val_loss: 4.3880 - val_acc: 0.0659
Epoch 5/10
- 23s - loss: 3.7760 - acc: 0.1298 - val_loss: 4.1772 - val_acc: 0.0814
Epoch 6/10
- 23s - loss: 3.4956 - acc: 0.1675 - val_loss: 4.3647 - val_acc: 0.0814
Epoch 7/10
- 23s - loss: 3.2150 - acc: 0.2249 - val_loss: 3.7449 - val_acc: 0.1353
Epoch 8/10
- 24s - loss: 2.9188 - acc: 0.2814 - val_loss: 3.6934 - val_acc: 0.1737
Epoch 9/10
- 23s - loss: 2.6100 - acc: 0.3488 - val_loss: 4.3043 - val_acc: 0.1281
Epoch 10/10
- 23s - loss: 2.3057 - acc: 0.4075 - val_loss: 3.6112 - val_acc: 0.1808

Out[61]:
<keras.callbacks.History at 0x2d24a33dc88>

### Data Augmentation¶

In [16]:
from keras.preprocessing.image import ImageDataGenerator
training_generator = ImageDataGenerator(rescale=1.,
rotation_range=15.0,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=0.1,
zoom_range=0.1,
horizontal_flip=True,
vertical_flip=False,
fill_mode="reflect")

validation_generator = ImageDataGenerator(rescale=1)

training_generator.fit(train_tensors)
validation_generator.fit(valid_tensors)

In [17]:
training_data = training_generator.flow(train_tensors, train_targets, batch_size=32)
validation_data = validation_generator.flow(valid_tensors, valid_targets, batch_size=32)

In [ ]:
model.load_weights('saved_models/weights.initial_scratch_model.hdf5')

In [66]:
from keras.callbacks import ModelCheckpoint

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

epochs = 10

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

checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch.hdf5',
verbose=0, save_best_only=True)

model.fit_generator(training_data, steps_per_epoch=train_tensors.shape[0]//32, epochs=epochs,
callbacks=[checkpointer], verbose=2,
validation_data=validation_data, validation_steps=valid_tensors.shape[0]//32)

Epoch 1/10
- 65s - loss: 4.7964 - acc: 0.0175 - val_loss: 4.9647 - val_acc: 0.0132
Epoch 2/10
- 64s - loss: 4.5376 - acc: 0.0394 - val_loss: 4.8624 - val_acc: 0.0156
Epoch 3/10
- 63s - loss: 4.3297 - acc: 0.0539 - val_loss: 4.5009 - val_acc: 0.0503
Epoch 4/10
- 65s - loss: 4.0879 - acc: 0.0823 - val_loss: 4.9490 - val_acc: 0.0503
Epoch 5/10
- 66s - loss: 3.7824 - acc: 0.1258 - val_loss: 4.3131 - val_acc: 0.0671
Epoch 6/10
- 63s - loss: 3.5273 - acc: 0.1560 - val_loss: 4.0201 - val_acc: 0.1353
Epoch 7/10
- 65s - loss: 3.3126 - acc: 0.2002 - val_loss: 3.7644 - val_acc: 0.1389
Epoch 8/10
- 64s - loss: 3.0704 - acc: 0.2417 - val_loss: 3.4313 - val_acc: 0.1940
Epoch 9/10
- 65s - loss: 2.8481 - acc: 0.2860 - val_loss: 4.0201 - val_acc: 0.1246
Epoch 10/10
- 65s - loss: 2.6400 - acc: 0.3237 - val_loss: 3.4074 - val_acc: 0.2132

Out[66]:
<keras.callbacks.History at 0x2d24a79d5f8>

### Load the Model with the Best Validation Loss¶

In [90]:
model.load_weights('saved_models/weights.best.from_scratch_no_da.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 [91]:
# 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]
dog_breed_predictions_DA = [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)

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

Test accuracy: 18.5407%
Test accuracy with Data Augmentation: 21.2919%


So adding Data Augmentation increases the accuracy, however the training time increases from ~ 23s to ~63s per epoch. The training accuracy using data augmentation is much less after 10 epochs, 32% vs 40% without data augmentation. This indicates that the model is more resistant to overtraining and generalizes better.

If the models were trained longer, the model with data augmentation would most likely continue to improve while the model without data augmentation would begin to overfit.

## 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('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 [21]:
VGG16_model = Sequential()

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 [22]:
VGG16_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])


### Train the Model¶

In [24]:
from keras.callbacks import ModelCheckpoint
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.udacityVGG16.hdf5',
verbose=0, 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=2)

Train on 6680 samples, validate on 835 samples
Epoch 1/20
- 4s - loss: 12.2185 - acc: 0.1174 - val_loss: 10.6527 - val_acc: 0.1928
Epoch 2/20
- 3s - loss: 9.8740 - acc: 0.2793 - val_loss: 9.9501 - val_acc: 0.2659
Epoch 3/20
- 3s - loss: 9.2901 - acc: 0.3470 - val_loss: 9.7107 - val_acc: 0.2982
Epoch 4/20
- 3s - loss: 8.9547 - acc: 0.3909 - val_loss: 9.4757 - val_acc: 0.3090
Epoch 5/20
- 3s - loss: 8.7490 - acc: 0.4127 - val_loss: 9.3520 - val_acc: 0.3365
Epoch 6/20
- 4s - loss: 8.5845 - acc: 0.4316 - val_loss: 9.1359 - val_acc: 0.3485
Epoch 7/20
- 4s - loss: 8.3560 - acc: 0.4442 - val_loss: 8.8905 - val_acc: 0.3617
Epoch 8/20
- 3s - loss: 8.1284 - acc: 0.4612 - val_loss: 8.6363 - val_acc: 0.3617
Epoch 9/20
- 3s - loss: 7.7975 - acc: 0.4783 - val_loss: 8.4271 - val_acc: 0.3808
Epoch 10/20
- 4s - loss: 7.6529 - acc: 0.5007 - val_loss: 8.4477 - val_acc: 0.3832
Epoch 11/20
- 3s - loss: 7.5986 - acc: 0.5105 - val_loss: 8.2982 - val_acc: 0.4024
Epoch 12/20
- 3s - loss: 7.5645 - acc: 0.5163 - val_loss: 8.4299 - val_acc: 0.4012
Epoch 13/20
- 3s - loss: 7.5483 - acc: 0.5192 - val_loss: 8.3252 - val_acc: 0.4000
Epoch 14/20
- 3s - loss: 7.5015 - acc: 0.5220 - val_loss: 8.2652 - val_acc: 0.4084
Epoch 15/20
- 3s - loss: 7.3434 - acc: 0.5319 - val_loss: 8.1379 - val_acc: 0.4216
Epoch 16/20
- 3s - loss: 7.2930 - acc: 0.5398 - val_loss: 8.1680 - val_acc: 0.4204
Epoch 17/20
- 2s - loss: 7.2790 - acc: 0.5404 - val_loss: 8.1560 - val_acc: 0.4180
Epoch 18/20
- 3s - loss: 7.2265 - acc: 0.5421 - val_loss: 8.0841 - val_acc: 0.4144
Epoch 19/20
- 3s - loss: 7.1718 - acc: 0.5476 - val_loss: 8.0289 - val_acc: 0.4311
Epoch 20/20
- 3s - loss: 7.0945 - acc: 0.5516 - val_loss: 7.9525 - val_acc: 0.4323

Out[24]:
<keras.callbacks.History at 0x17037222dd8>

### Load the Model with the Best Validation Loss¶

In [25]:
VGG16_model.load_weights('saved_models/weights.best.udacityVGG16.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 [26]:
# 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: 43.4211%


### Predict Dog Breed with the Model¶

In [27]:
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:

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. Pick one of the above architectures, download the corresponding bottleneck features, and store the downloaded file in the bottleneck_features/ folder in the repository.

### (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('bottleneck_features/Dog{network}Data.npz')
train_{network} = bottleneck_features['train']
valid_{network} = bottleneck_features['valid']
test_{network} = bottleneck_features['test']
In [28]:
### TODO: Obtain bottleneck features from another pre-trained CNN.

train_InceptionV3 = bottleneck_features['train']
valid_InceptionV3 = bottleneck_features['valid']
test_InceptionV3 = 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: I kept the same top model as my scratch model and used the Inception model as a feature extractor. I played around with more complicated networks, but this simple network seemed to work just as well. I extended this model architecture to the other pretrained models and also tried data augmentation. I then organized the best models into an ensemble. Below, I have the basic Inception model and then show the better model of the ensemble of different models. The process of reaching that was pretty involved so I organized it in a separate notebook which can be viewed here.

This architecture is suitable because it outperforms all the other models. Specifically, the use of pretrained networks to extract the features of the images drastically improves the accuracy of the model since the pretrained models were trained on a much larger, much more diverse training set, for much longer. The GAP layer feeding directly to the output nodes also allows for the visualization of what the model is seeing. This is touched on briefly in the additional notebook mentioned above. Additionally, the small models make inference and training faster.

In [39]:
### TODO: Define your architecture.

inception_model = Sequential()


In [33]:
inception_model.summary()

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
global_average_pooling2d_3 ( (None, 2048)              0
_________________________________________________________________
batch_normalization_96 (Batc (None, 2048)              8192
_________________________________________________________________
dense_3 (Dense)              (None, 133)               272517
=================================================================
Total params: 280,709
Trainable params: 276,613
Non-trainable params: 4,096
_________________________________________________________________


### (IMPLEMENTATION) Compile the Model¶

In [40]:
### TODO: Compile the model.
from keras.optimizers import Adam


### (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 [41]:
### TODO: Train the model.
from keras.callbacks import ModelCheckpoint
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.projectinception.hdf5',
verbose=0, save_best_only=True)

inception_model.fit(train_InceptionV3, train_targets,
validation_data=(valid_InceptionV3, valid_targets),
epochs=10, batch_size=32, callbacks=[checkpointer], verbose=2)

Train on 6680 samples, validate on 835 samples
Epoch 1/10
- 5s - loss: 1.0959 - acc: 0.7211 - val_loss: 0.6732 - val_acc: 0.8168
Epoch 2/10
- 3s - loss: 0.3098 - acc: 0.9010 - val_loss: 0.6499 - val_acc: 0.8299
Epoch 3/10
- 3s - loss: 0.1586 - acc: 0.9491 - val_loss: 0.6511 - val_acc: 0.8359
Epoch 4/10
- 3s - loss: 0.0979 - acc: 0.9702 - val_loss: 0.6408 - val_acc: 0.8431
Epoch 5/10
- 3s - loss: 0.0666 - acc: 0.9807 - val_loss: 0.6620 - val_acc: 0.8383
Epoch 6/10
- 4s - loss: 0.0455 - acc: 0.9868 - val_loss: 0.6170 - val_acc: 0.8503
Epoch 7/10
- 4s - loss: 0.0353 - acc: 0.9916 - val_loss: 0.6688 - val_acc: 0.8479
Epoch 8/10
- 4s - loss: 0.0290 - acc: 0.9934 - val_loss: 0.6346 - val_acc: 0.8599
Epoch 9/10
- 4s - loss: 0.0331 - acc: 0.9918 - val_loss: 0.7183 - val_acc: 0.8419
Epoch 10/10
- 4s - loss: 0.0290 - acc: 0.9945 - val_loss: 0.6853 - val_acc: 0.8455

Out[41]:
<keras.callbacks.History at 0x17444e25eb8>

### (IMPLEMENTATION) Load the Model with the Best Validation Loss¶

In [42]:
### TODO: Load the model weights with the best validation loss.



### (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 [43]:
### TODO: Calculate classification accuracy on the test dataset.
# get index of predicted dog breed for each image in test set
inception_predictions = [np.argmax(inception_model.predict(np.expand_dims(feature, axis=0))) for feature in test_InceptionV3]

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

Test accuracy: 81.8182%


### Better Model¶

The process in reaching this model can be found in this notebook which organizes a lot of experimentation that went into reaching this model.

#### Data generation¶

In [51]:
from keras.utils.np_utils import to_categorical
inception_generator = ImageDataGenerator(rescale=1/255)
test_data_path = 'dogImages/test/'
batch_size = 32

In [52]:
inception_test_data = inception_generator.flow_from_directory(test_data_path,
target_size=(350, 350),
batch_size=batch_size,
shuffle=False,
class_mode='categorical')

test_samples = len(inception_test_data.filenames)
test_labels = to_categorical(inception_test_data.classes)
dog_names = [item[20:-1] for item in sorted(glob("dogImages/train/*/"))]

Found 836 images belonging to 133 classes.


#### Inception Model¶

In [53]:
from keras.applications.inception_v3 import InceptionV3
inception = InceptionV3(include_top=False, weights='imagenet')

inception_model = Sequential()



#### Xception Model¶

In [54]:
from keras.applications.xception import Xception, preprocess_input
xception = Xception(include_top=False, weights='imagenet')

xception_model = Sequential()



#### InceptionResnet¶

In [55]:
from keras.applications.inception_resnet_v2 import InceptionResNetV2, preprocess_input
inceptionresnet = InceptionResNetV2(include_top=False, weights='imagenet')

inceptionresnet_model = Sequential()


In [56]:
inception_features = inception.predict_generator(inception_test_data, steps=1)
xception_features = xception.predict_generator(inception_test_data, steps=1)
inceptionresnet_features = inceptionresnet.predict_generator(inception_test_data, steps=1)

In [57]:
inception_weights = [
'saved_models/weights.best.inception.hdf5',
'saved_models/weights.best.inception2.hdf5',
'saved_models/weights.best.inception3.hdf5',
'saved_models/weights.best.inception4.hdf5',
'saved_models/weights.best.inception5.hdf5',
'saved_models/weights.best.inception6.hdf5'
]

xception_weights = [
'saved_models/weights.best.xception.hdf5',
'saved_models/weights.best.xception2.hdf5',
'saved_models/weights.best.xception3.hdf5',
'saved_models/weights.best.xception4.hdf5',
'saved_models/weights.best.xception5.hdf5',
'saved_models/weights.best.xception6.hdf5'
]

inceptionresnet_weights = [
'saved_models/weights.best.inceptionresnet.hdf5',
'saved_models/weights.best.inceptionresnet2.hdf5',
'saved_models/weights.best.inceptionresnet3.hdf5',
'saved_models/weights.best.inceptionresnet4.hdf5',
'saved_models/weights.best.inceptionresnet5.hdf5',
'saved_models/weights.best.inceptionresnet6.hdf5'
]

predictions = np.zeros((test_samples, 133), dtype=np.float)

for weight_path in inception_weights:
model_prediction = inception_model.predict(inception_features)
predictions += model_prediction
for weight_path in xception_weights:
model_prediction = xception_model.predict(xception_features)
predictions += model_prediction
for weight_path in inceptionresnet_weights:
model_prediction = inceptionresnet_model.predict(inceptionresnet_features)
predictions += model_prediction

predictions /= 18

In [60]:
import keras
import keras.backend as K
scores = K.eval(keras.metrics.categorical_accuracy(test_labels, predictions))
test_accuracy = 100*scores.mean()
print('Test accuracy: %.4f%%' % test_accuracy)

Test accuracy: 93.6603%


### (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 [159]:
### 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 path_to_tensor350(img_path):
# loads RGB image as PIL.Image.Image type
img = image.load_img(img_path, target_size=(350, 350))
# 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 ensemble_predict_breed(img_path):
# Get bottleneck features
inception_features = inception.predict(path_to_tensor350(img_path)/255)
xception_features = xception.predict(path_to_tensor350(img_path)/255)
inceptionresnet_features = inceptionresnet.predict(path_to_tensor350(img_path)/255)

predictions = np.zeros((1, 133), dtype=np.float)

for weight_path in inception_weights:
model_prediction = inception_model.predict(inception_features)
predictions += model_prediction
for weight_path in xception_weights:
model_prediction = xception_model.predict(xception_features)
predictions += model_prediction
for weight_path in inceptionresnet_weights:
model_prediction = inceptionresnet_model.predict(inceptionresnet_features)
predictions += model_prediction

predictions /= 18

value_dict = {value:idx for idx,value in enumerate(predictions[0])}
top_breeds = list(value_dict.keys())
top_breeds.sort()

# Get sum of top 5 so the percentages can be normalized to only the top 5 breed predictions
sum_of_top5 = sum(top_breeds[-5::])

print('Our prediction for your breed is:')
for i in range(5):
value = top_breeds[-(i+1)]
breed = dog_names[value_dict[value]]
print(str(round(100*value/sum_of_top5, 2)) + '% ' + breed)


## 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 [157]:
### TODO: Write your algorithm.
### Feel free to use as many code cells as needed.

# Get image

# Is there a human

# Is there a dog
def breed_classifier(img_path):
'''
Takes image, checks if there is a human or a dog. Predicts breed.
'''

human_present = face_detector(img_path)
dog_present = dog_detector(img_path)

if human_present and dog_present:
print('A dog and a human were detected in the image. Try a different image with only a human or a dog.')

elif not(human_present or dog_present):
print('A dog or person could not be detected from the provided image. Please use another.')
else:
if human_present:
print('Hello, human!')
else:
print('Hello, dog')
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
plt.imshow(cv_rgb)
plt.show()
ensemble_predict_breed(img_path)



Extra ideas:

1) Train model on augmented training data

2) train last few layers of CNN

3) Rather than outputing the breed, output the top 2-3 classes and the percentages

4) Use an ensemble of networks for breed classification

5) object detection algorithm - check out fast.ai lesson 9; the linked project and karpathy's course

Better human/face detector

5) For humans you can output a picture of the top predicted dog breeds, ideally one with similar activations

## 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.

In [158]:
## TODO: Execute your algorithm from Step 6 on
## at least 6 images on your computer.
## Feel free to use as many code cells as needed.

breed_classifier(human_files_short[3])

Hello, human!

Our prediction for your breed is:
37.26% Dogue_de_bordeaux
19.95% Afghan_hound
17.35% Lowchen
13.5% Dachshund
11.94% Bulldog

In [164]:
breed_classifier(human_files_short[4])

Hello, human!

Our prediction for your breed is:
28.13% Smooth_fox_terrier
26.35% Finnish_spitz
19.39% Dogue_de_bordeaux
13.36% Portuguese_water_dog
12.77% Dachshund

In [174]:
print(dog_files_short[5])
breed_classifier(dog_files_short[5])

dogImages/train\063.English_springer_spaniel\English_springer_spaniel_04469.jpg
Hello, dog

Our prediction for your breed is:
94.89% English_springer_spaniel
2.74% Irish_red_and_white_setter
1.12% Welsh_springer_spaniel
0.74% English_cocker_spaniel
0.5% Brittany

In [180]:
print(dog_files_short[10])
breed_classifier(dog_files_short[10])

dogImages/train\012.Australian_shepherd\Australian_shepherd_00814.jpg
Hello, dog

Our prediction for your breed is:
99.83% Australian_shepherd
0.07% Border_collie
0.05% Icelandic_sheepdog
0.03% Nova_scotia_duck_tolling_retriever
0.02% Cardigan_welsh_corgi

In [188]:
print(dog_files_short[20])
breed_classifier(dog_files_short[20])

dogImages/train\097.Lakeland_terrier\Lakeland_terrier_06526.jpg
Hello, dog