

In [1]:

    
%matplotlib inline
%load_ext autoreload
%autoreload 2

from __future__ import print_function

import os
import sys
import matplotlib.pyplot as plt

from keras.applications.imagenet_utils import preprocess_input
from keras.models import load_model
from keras.datasets import mnist

sys.path.append('..')
from utils import display_examples, RotNetDataGenerator, angle_error, binarize_images









    



Using TensorFlow backend.

Load data



In [2]:

    
(X_train, y_train), (X_test, y_test) = mnist.load_data()

Load model



In [3]:

    
model_location = os.path.join('..', 'models', 'rotnet_mnist.hdf5')
model = load_model(model_location, custom_objects={'angle_error': angle_error})

Evaluate model



In [4]:

    
batch_size = 128
out = model.evaluate_generator(
    RotNetDataGenerator(
        X_test,
        batch_size=batch_size,
        preprocess_func=binarize_images,
        shuffle=True
    ), 
    steps=len(y_test) / batch_size
)

print('Test loss:', out[0])
print('Test angle error:', out[1])









    



Test loss: 2.68059594193
Test angle error: 6.4525

Display examples



In [5]:

    
num_images = 5

display_examples(
    model, 
    X_test,
    num_images=num_images,
    preprocess_func=binarize_images,
)