Joshua SierlesIdentifying Image Subjects with Keras, Our Version
Classify images with a Keras/Tensorflow Convolutional Neural Network using the CIFAR-10 image dataset
Keras is a high-level Python neural network API, designed for ease-of-use multiple backend support. The default backend is Tensorflow. We'll use Tensorflow to classify the CIFAR-10 image dataset.
1. Setup
Since we'll run this training on a GPU, we'll need tensorflow-gpu and the CUDA neural network libraries. We need h5py for storing the training results in HD5 format. Let's also add graphical packages for visualizing our training models.
conda install -qy -c anaconda tensorflow-gpu h5py graphviz pydot pip install keras dill
conda install -cy tensorflow-gpu=1.2.1
The CIFAR-10 dataset is a collection of 60,000 color, 32x32-pixel images in ten classes, 10,000 of which are in a test batch. Keras can automatically download the dataset, but let's save some time by doing that now—it will cache the download, and we'll lock this cell and the setup cell above after the first run so that we shouldn't need to redownload.
from keras.datasets import cifar10 cifar10.load_data()
We'll run 128-image batches and set up two training runs: a long, 500-epoch run to do the main work, and a short, 5-epoch run as an example.
batch_size = 128 num_classes = 10 epochs_shortrun = 5 epochs_longrun = 500 save_dir = "/work" res_dir = "/results" model_name = 'convnet_cifar10'
Load the data and get it into a reasonable shape. Also set up a function to find the best checkpoint file, another to give us a look at the images we're analyzing, and finally set up to do real-time input-data augmentation.
from __future__ import print_function import tensorflow as tf import keras from keras.datasets import cifar10 from keras.preprocessing.image import ImageDataGenerator from keras.models import Sequential from keras.layers.core import Dense, Dropout, Flatten from keras.layers.convolutional import Conv2D from keras.optimizers import Adam from keras.layers.pooling import MaxPooling2D from keras.callbacks import ModelCheckpoint,EarlyStopping from keras.utils import to_categorical from keras.models import load_model import os import dill as pickle import numpy as np # set random seeds for reproducibility tf.reset_default_graph() tf.set_random_seed(343) np.random.seed(343) os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2" # The data, shuffled and split between train and test sets: (x_train, y_train), (x_test, y_test) = cifar10.load_data() print('x_train shape:', x_train.shape) print(x_train.shape[0], 'train samples') print(x_test.shape[0], 'test samples') # Convert class vectors to binary class matrices. y_train = to_categorical(y_train, num_classes) y_test = to_categorical(y_test, num_classes) x_train = x_train.astype('float32') x_test = x_test.astype('float32') x_train /= 255.0 x_test /= 255.0 # Load label names to use in prediction results label_list_path = 'datasets/cifar-10-batches-py/batches.meta' keras_dir = os.path.expanduser(os.path.join('~', '.keras')) datadir_base = os.path.expanduser(keras_dir) if not os.access(datadir_base, os.W_OK): datadir_base = os.path.join('/tmp', '.keras') label_list_path = os.path.join(datadir_base, label_list_path) with open(label_list_path, mode='rb') as f: labels = pickle.load(f) ckpt_dir = os.path.join(save_dir,"checkpoints") if not os.path.isdir(ckpt_dir): os.makedirs(ckpt_dir) model_path = os.path.join(res_dir, model_name + ".kerasave") hist_path = os.path.join(res_dir, model_name + ".kerashist") # Function to find latest checkpoint file def last_ckpt(dir): fl = os.listdir(dir) fl = [x for x in fl if x.endswith(".hdf5")] cf = "" if len(fl) > 0: accs = [float(x.split("-")[3][0:-5]) for x in fl] m = max(accs) iaccs = [i for i, j in enumerate(accs) if j == m] fl = [fl[x] for x in iaccs] epochs = [int(x.split("-")[2]) for x in fl] cf = fl[epochs.index(max(epochs))] cf = os.path.join(dir,cf) return(cf) import matplotlib.pyplot as plt from math import * #Visualizing CIFAR 10, takes indicides and shows in a grid def cifar_grid(X,Y,inds,n_col, predictions=None): if predictions is not None: if Y.shape != predictions.shape: print("Predictions must equal Y in length!") return(None) N = len(inds) n_row = int(ceil(1.0*N/n_col)) fig, axes = plt.subplots(n_row,n_col,figsize=(10,10)) clabels = labels["label_names"] for j in range(n_row): for k in range(n_col): i_inds = j*n_col+k i_data = inds[i_inds] axes[j][k].set_axis_off() if i_inds < N: axes[j][k].imshow(X[i_data,...], interpolation='nearest') label = clabels[np.argmax(Y[i_data,...])] axes[j][k].set_title(label) if predictions is not None: pred = clabels[np.argmax(predictions[i_data,...])] if label != pred: label += " n" axes[j][k].set_title(pred, color='red') fig.set_tight_layout(True) return(fig) print('Using real-time data augmentation.') # This will do preprocessing and realtime data augmentation: datagen = ImageDataGenerator( featurewise_center=False, # set input mean to 0 over the dataset samplewise_center=False, # set each sample mean to 0 featurewise_std_normalization=False, # divide inputs by std of the dataset samplewise_std_normalization=False, # divide each input by its std zca_whitening=False, # apply ZCA whitening rotation_range=0, # randomly rotate images in the range (degrees, 0 to 180) width_shift_range=0.1, # randomly shift images horizontally (fraction of total width) height_shift_range=0.1, # randomly shift images vertically (fraction of total height) horizontal_flip=True, # randomly flip images vertical_flip=False) # randomly flip images # Compute quantities required for feature-wise normalization # (std, mean, and principal components if ZCA whitening is applied). datagen.fit(x_train)
Let's take a gander at a random selection of training images.
indices = [np.random.choice(range(len(x_train))) for i in range(36)] cifar_grid(x_train,y_train,indices,6)
We'll use a simple convolutional network model (still under development), with the addition of the data augmentation defined above, and a checkpoint-writing callback that's keyed to significant accuracy improvements.
model = Sequential() model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=x_train.shape[1:])) model.add(Conv2D(64, kernel_size=(3, 3), activation='relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.25)) model.add(Conv2D(128, kernel_size=(3, 3), activation='relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Conv2D(128, kernel_size=(3, 3), activation='relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.25)) model.add(Flatten()) model.add(Dense(1024, activation='relu')) model.add(Dropout(0.5)) model.add(Dense(10, activation='softmax')) # initiate Adam optimizer opt = Adam(lr=0.0001, decay=1e-6) # Let's train the model using RMSprop model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy']) # checkpoint callback filepath = os.path.join(ckpt_dir, "weights-improvement-{epoch:02d}-{val_acc:.6f}.hdf5") checkpoint = ModelCheckpoint( filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='max' ) print("Saving improvement checkpoints to \n\t{0}".format(filepath)) # early stop callback, given a bit more leeway stahp = EarlyStopping(min_delta=0.000001, patience=25)
Finally, let's take a look at our model, with both a text summary and a flow chart.
from keras.utils import plot_model plot_model(model, to_file="/results/model.svg", show_layer_names=False, show_shapes=True, rankdir="TB")
2. Training
Now we're ready to train using the GPU. We'll set up two code branches: one to do some serious long-term training (takes hours), and one which just runs a few additional epochs as an example. Note that both act as endpoints to their inheritance branches, because the underlying system that allows inheritance does not currenly know how to handle GPU activity. So, the training cells will save their results to /files/, and then for analysis and visualization we'll just need to load that data. We'll also pickle the training history for the long run to a file, in case we want to take a look at that.
Both paths are able to load the most accurate of any existing checkpoints in /files/checkpoints—this would allow for additional refinement of accuracy, although such runs would probably require removal or relaxation of the EarlyStopping callback.
2.1. Long Training
epochs = settings.epochs_longrun cpf = last_ckpt(ckpt_dir) if cpf != "": print("Loading starting weights from \n\t{0}".format(cpf)) model.load_weights(cpf) # Fit the model on the batches generated by datagen.flow(). hist = model.fit_generator(datagen.flow(x_train, y_train, batch_size=batch_size,shuffle=True), steps_per_epoch=x_train.shape[0] // batch_size, epochs=epochs,verbose=2, validation_data=(x_test, y_test), workers=4, callbacks=[checkpoint,stahp]) # Save model and weights model.save(model_path) #print('Saved trained model at %s ' % model_path) with open(hist_path, 'wb') as f: pickle.dump(hist.history, f)
2.2. Short Example
epochs = settings.epochs_shortrun # load results of long training run model.load_weights(long training.convnet_cifar10.kerasave) # Fit the model on the batches generated by datagen.flow(). hist = model.fit_generator(datagen.flow(x_train, y_train, batch_size=batch_size), steps_per_epoch=x_train.shape[0] // batch_size, epochs=epochs,verbose=2, validation_data=(x_test, y_test), workers=4, callbacks=[checkpoint]) # Save model and weights model.save(model_path) print('Saved trained model at %s ' % model_path)
3. Results
Alrighty, now we can take a look at the trained model. The load_model() function will give us back our full, trained model for evaluation and prediction.
model = load_model(short training.convnet_cifar10.kerasave) # Evaluate model with test data set evaluation = model.evaluate_generator(datagen.flow(x_test, y_test, batch_size=batch_size, shuffle=False), steps=x_test.shape[0] // batch_size, workers=4) # Print out final values of all metrics key2name = {'acc':'Accuracy', 'loss':'Loss', 'val_acc':'Validation Accuracy', 'val_loss':'Validation Loss'} results = [] for i,key in enumerate(model.metrics_names): results.append('%s = %.2f' % (key2name[key], evaluation[i])) print(", ".join(results))
We can sample the prediction results with images.
num_predictions = 36 model = load_model(short training.convnet_cifar10.kerasave) predict_gen = model.predict_generator(datagen.flow(x_test, y_test, batch_size=batch_size, shuffle=False), steps=(x_test.shape[0] // batch_size)+1, workers=4) indices = [np.random.choice(range(len(x_test))) for i in range(num_predictions)] cifar_grid(x_test,y_test,indices,4, predictions=predict_gen)
And hey, let's take a look at the training history (we'll look at the long training so it's an interesting history).
import matplotlib import matplotlib.pyplot as plt with open(long training.convnet_cifar10.kerashist, 'rb') as f: hist = pickle.load(f) key2name = {'acc':'Accuracy', 'loss':'Loss', 'val_acc':'Validation Accuracy', 'val_loss':'Validation Loss'} fig = plt.figure() things = ['acc','loss','val_acc','val_loss'] for i,thing in enumerate(things): trace = hist[thing] plt.subplot(2,2,i+1) plt.plot(range(len(trace)),trace) plt.title(key2name[thing]) fig.set_tight_layout(True) fig
Finally, let's see if our trained network can correctly identify the subject of an uploaded image. This is the Internet, so it must be a cat.
This might be a cat
from keras import backend as K sess = K.get_session() model = load_model(short training.convnet_cifar10.kerasave) img = tf.read_file(horse.jpg) img = tf.image.decode_jpeg(img, channels=3) img.set_shape([None, None, 3]) img = tf.image.resize_images(img, (32, 32)) img = img.eval(session=sess) # convert to numpy array img = np.expand_dims(img, 0) # make 'batch' of 1 pred = model.predict(img) pred = labels["label_names"][np.argmax(pred)]
Magic 8-ball says this image contains a horse. Huzzah!