This is a repo for the Dog Breed Classifier Project in Udacity Nanodegree Consist of 2 parts:
- model training
- inference web app It is implemented by using numpy, Keras, tensorflow, flask
Udacity's original repo is here
Welcome to the Convolutional Neural Networks (CNN) project in the AI Nanodegree! In this project, you will learn how to build a pipeline that can be used within a web or mobile app to process real-world, user-supplied images. Given an image of a dog, your algorithm will identify an estimate of the canine’s breed. If supplied an image of a human, the code will identify the resembling dog breed.
Along with exploring state-of-the-art CNN models for classification and localization, you will make important design decisions about the user experience for your app. Our goal is that by completing this lab, you understand the challenges involved in piecing together a series of models designed to perform various tasks in a data processing pipeline. Each model has its strengths and weaknesses, and engineering a real-world application often involves solving many problems without a perfect answer. Your imperfect solution will nonetheless create a fun user experience!
Generate by running: $ tree --charset=ascii
|-- dog_breed_image_classifier_train
| |-- README.md
| |-- data
| |-- dog_app.html
| |-- dog_app.ipynb
| |-- images
| | |-- Brittany_02625.jpg
| | |-- sample_human_2.png
| |-- requirements.txt
| `-- saved_models
| |-- weights.best.VGG16.hdf5
| |-- weights.best.Xception.hd5
| `-- weights.best.from_scratch.hdf5
dog_app.ipynb is the main code to train the model. Models are saved under saved_models. You can check dog_app.html for the result.
- Kaggle Compitition
https://www.kaggle.com/c/dog-breed-identification - Download the dog dataset
wget https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/dogImages.zip - Download the human dataset
wget https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/lfw.zip
Model: "model_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_2 (InputLayer) (None, 224, 224, 3) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 223, 223, 16) 208
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 111, 111, 16) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 110, 110, 32) 2080
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 55, 55, 32) 0
_________________________________________________________________
conv2d_3 (Conv2D) (None, 54, 54, 64) 8256
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 27, 27, 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
_________________________________________________________________
After trained for 20 epochs, training accuracy is 5.76%, testing accuracy is only 5.86%.
Model: "model_2"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_3 (InputLayer) (None, 224, 224, 3) 0
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, 224, 224, 4) 304 input_3[0][0]
__________________________________________________________________________________________________
max_pooling2d_5 (MaxPooling2D) (None, 112, 112, 4) 0 conv2d_4[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, 56, 56, 4) 148 max_pooling2d_5[0][0]
__________________________________________________________________________________________________
conv2d_6 (Conv2D) (None, 56, 56, 8) 40 conv2d_5[0][0]
__________________________________________________________________________________________________
conv2d_8 (Conv2D) (None, 56, 56, 8) 40 conv2d_5[0][0]
__________________________________________________________________________________________________
max_pooling2d_6 (MaxPooling2D) (None, 56, 56, 4) 0 conv2d_5[0][0]
__________________________________________________________________________________________________
conv2d_7 (Conv2D) (None, 56, 56, 8) 584 conv2d_6[0][0]
__________________________________________________________________________________________________
conv2d_9 (Conv2D) (None, 56, 56, 8) 1608 conv2d_8[0][0]
__________________________________________________________________________________________________
conv2d_10 (Conv2D) (None, 56, 56, 8) 40 max_pooling2d_6[0][0]
__________________________________________________________________________________________________
concatenate_1 (Concatenate) (None, 56, 56, 24) 0 conv2d_7[0][0]
conv2d_9[0][0]
conv2d_10[0][0]
__________________________________________________________________________________________________
max_pooling2d_7 (MaxPooling2D) (None, 28, 28, 24) 0 concatenate_1[0][0]
__________________________________________________________________________________________________
conv2d_11 (Conv2D) (None, 28, 28, 4) 100 max_pooling2d_7[0][0]
__________________________________________________________________________________________________
flatten_1 (Flatten) (None, 3136) 0 conv2d_11[0][0]
__________________________________________________________________________________________________
dense_2 (Dense) (None, 133) 417221 flatten_1[0][0]
==================================================================================================
Total params: 420,085
Trainable params: 420,085
Non-trainable params: 0
After trained for 20 epochs, training accuracy is 99.45%, testing accuracy is only 5.38%. The model is overfitting.
The model uses the the pre-trained model as a fixed feature extractor, where the last convolutional output 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.
def transfer_learning(network):
if network =='ResNet-50':
bottleneck_features_network = np.load('data/bottleneck_features/DogResnet50Data.npz')
elif network == 'InceptionV3':
bottleneck_features_network = np.load('data/bottleneck_features/DogInceptionV3Data.npz')
elif network =='Xception':
bottleneck_features_network = np.load('data/bottleneck_features/DogXceptionData.npz')
elif network =='VGG-16':
bottleneck_features_network = np.load('data/bottleneck_features/DogVGG16Data.npz')
elif network =='VGG-19':
bottleneck_features_network = np.load('data/bottleneck_features/DogVGG19Data.npz')
train_network = bottleneck_features_network['train']
valid_network = bottleneck_features_network['valid']
test_network = bottleneck_features_network['test']
### Define your architecture.
Network_model = Sequential()
Network_model.add(GlobalAveragePooling2D(input_shape=train_network.shape[1:]))
Network_model.add(Dense(133, activation='softmax'))
### Compile the model.
Network_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
### Train the model.
checkpointer = ModelCheckpoint(filepath=model_filepath, verbose=1, save_best_only=True)
## Remember --> train_targets, valid_targets, test_targets contains the one-hot encoded correct values.
Network_model.fit(train_network, train_targets,
validation_data=(valid_network, valid_targets),
epochs=10, batch_size=1, callbacks=[checkpointer], verbose=1)
### Load the model weights with the best validation loss.
Network_model.load_weights(model_filepath)
print("The best weights for the {} model have been loaded".format(network))
### Calculate classification accuracy on the test dataset.
predictions = [np.argmax(Network_model.predict(np.expand_dims(feat, axis=0))) for feat in test_network]
test_accuracy = 100*np.sum(np.array(predictions)==np.argmax(test_targets, axis=1))/len(predictions)
print('Test accuracy: %.4f%%' % test_accuracy)Our model:
Model: "sequential_4"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
global_average_pooling2d_4 ( (None, 512) 0
_________________________________________________________________
dense_5 (Dense) (None, 133) 68229
=================================================================
Total params: 68,229
Trainable params: 68,229
Non-trainable params: 0
Transfer learning bottleneck features
* RestNet50: 77M
wget https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/DogResnet50Data.npz
* VGG16: 811M
wget https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/DogVGG16Data.npz
* VGG19: 811M
wget https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/DogVGG19Data.npz
* Xception: 3.1G
wget https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/DogXceptionData.npz
* InceptionV3: 1.6G
https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/DogInceptionV3Data.npz
Results:
- transfer_learning('VGG-19') After trained for 20 epochs, training accuracy is 83.68%, testing accuracy is up to 67.10%.
- transfer_learning('VGG-16') After trained for 20 epochs, training accuracy is 99.24%, testing accuracy is up to 70.93%.
- transfer_learning('ResNet-50') After trained for 20 epochs, training accuracy is 88.28%, testing accuracy is up to 77.63%.
- transfer_learning('Inception') After trained for 20 epochs, training accuracy is 87.60%, testing accuracy is up to 78.59%.
- transfer_learning('Xception') After trained for 20 epochs, training accuracy is 87.60%, testing accuracy is up to 84.21%.
Conclusion: ResNet50 achive pretty good result with less features file size as 77M. Xception achieve the best accuracy, but feature size is 3.1G.
Generate by running: $ tree --charset=ascii
`-- keras-flask-deploy-webapp
|-- Dockerfile
|-- README.md
|-- app.py
|-- build.sh
|-- data
| `-- dogImages
|-- dog_breed.py
|-- dog_breed_app.py
|-- models
| |-- weights.best.VGG16.hdf5
| |-- weights.best.VGG19.hdf5
| |-- weights.best.Xception.hd5
| `-- weights.best.from_scratch.hdf5
|-- requirements.txt
|-- run.sh
|-- static
| |-- css
| | `-- main.css
| `-- js
| `-- main.js
|-- templates
| |-- base.html
| `-- index.html
`-- uploads
`-- README.md
👇Screenshot:
$ cd keras-flask-deploy-webapp
$ docker build -t dog_breed_classifier
$ docker run --name dog_app --detach -p 3000:3000 dog_breed_classifierFor your convenience, can just pull the image instead of building it.
$ docker pull docker.io/kissingurami/dog_breed_classifier:1.0
$ docker run --name dog_app --detach -p 3000:3000 kissingurami/dog_breed_classifier:1.0
$ docker logs -f dog_appOpen http://localhost:3000 after waiting for a minute to install in the container.