util.py 9.68 KB
Newer Older
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
1
2
3
#!/usr/bin/env python
# vim: set fileencoding=utf-8 :
# @author: Tiago de Freitas Pereira <tiago.pereira@idiap.ch>
4
# @date: Wed 11 May 2016 09:39:36 CEST
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
5
6
7

import numpy
import tensorflow as tf
8
from tensorflow.python.client import device_lib
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
9

10

Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
11
12
13
14
15
def compute_euclidean_distance(x, y):
    """
    Computes the euclidean distance between two tensorflow variables
    """

16
    with tf.name_scope('euclidean_distance'):
17
        d = tf.sqrt(tf.reduce_sum(tf.square(tf.subtract(x, y)), 1))
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
18
19
        return d

20
21

def load_mnist(perc_train=0.9):
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
22
23

    import bob.db.mnist
24
    db = bob.db.mnist.Database()
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
25
26
27
28
29
30
    raw_data = db.data()

    # data  = raw_data[0].astype(numpy.float64)
    data = raw_data[0]
    labels = raw_data[1]

31
32
33
34
35
36
    # Shuffling
    total_samples = data.shape[0]
    indexes = numpy.array(range(total_samples))
    numpy.random.shuffle(indexes)

    # Spliting train and validation
37
    n_train = int(perc_train * indexes.shape[0])
38
39
    n_validation = total_samples - n_train

40
    train_data = data[0:n_train, :].astype("float32") * 0.00390625
41
42
    train_labels = labels[0:n_train]

43
44
45
    validation_data = data[n_train:n_train +
                           n_validation, :].astype("float32") * 0.00390625
    validation_labels = labels[n_train:n_train + n_validation]
46
47

    return train_data, train_labels, validation_data, validation_labels
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
48
49


50
def create_mnist_tfrecord(tfrecords_filename, data, labels, n_samples=6000):
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
51

52
53
    def _bytes_feature(value):
        return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
54

55
56
    def _int64_feature(value):
        return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
57

58
    writer = tf.python_io.TFRecordWriter(tfrecords_filename)
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
59

60
61
62
    for i in range(n_samples):
        img = data[i]
        img_raw = img.tostring()
63
64
        feature = {'data': _bytes_feature(img_raw),
                   'label': _int64_feature(labels[i]),
65
                   'key': _bytes_feature(b'-')
66
67
                   }

68
69
70
        example = tf.train.Example(features=tf.train.Features(feature=feature))
        writer.write(example.SerializeToString())
    writer.close()
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
71
72


73
74
def compute_eer(data_train, labels_train, data_validation, labels_validation,
                n_classes):
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
    import bob.measure
    from scipy.spatial.distance import cosine

    # Creating client models
    models = []
    for i in range(n_classes):
        indexes = labels_train == i
        models.append(numpy.mean(data_train[indexes, :], axis=0))

    # Probing
    positive_scores = numpy.zeros(shape=0)
    negative_scores = numpy.zeros(shape=0)

    for i in range(n_classes):
        # Positive scoring
        indexes = labels_validation == i
        positive_data = data_validation[indexes, :]
92
93
        p = [cosine(models[i], positive_data[j])
             for j in range(positive_data.shape[0])]
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
94
95
96
97
98
        positive_scores = numpy.hstack((positive_scores, p))

        # negative scoring
        indexes = labels_validation != i
        negative_data = data_validation[indexes, :]
99
100
        n = [cosine(models[i], negative_data[j])
             for j in range(negative_data.shape[0])]
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
101
102
103
104
105
106
107
108
109
110
111
112
113
        negative_scores = numpy.hstack((negative_scores, n))

    # Computing performance based on EER
    negative_scores = (-1) * negative_scores
    positive_scores = (-1) * positive_scores

    threshold = bob.measure.eer_threshold(negative_scores, positive_scores)
    far, frr = bob.measure.farfrr(negative_scores, positive_scores, threshold)
    eer = (far + frr) / 2.

    return eer


114
115
def compute_accuracy(data_train, labels_train, data_validation,
                     labels_validation, n_classes):
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
116
117
118
119
120
121
122
123
124
125
126
127
    from scipy.spatial.distance import cosine

    # Creating client models
    models = []
    for i in range(n_classes):
        indexes = labels_train == i
        models.append(numpy.mean(data_train[indexes, :], axis=0))

    # Probing
    tp = 0
    for i in range(data_validation.shape[0]):

128
        d = data_validation[i, :]
Tiago de Freitas Pereira's avatar
Scratch  
Tiago de Freitas Pereira committed
129
130
131
132
133
134
135
136
137
        l = labels_validation[i]

        scores = [cosine(m, d) for m in models]
        predict = numpy.argmax(scores)

        if predict == l:
            tp += 1

    return (float(tp) / data_validation.shape[0]) * 100
138
139


Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
140
141
142
143
144
def debug_embbeding(image, architecture, embbeding_dim=2, feature_layer="fc3"):
    """
    """
    import tensorflow as tf
    from bob.learn.tensorflow.utils.session import Session
145
146
147
148
149

    session = Session.instance(new=False).session
    inference_graph = architecture.compute_graph(
        architecture.inference_placeholder, feature_layer=feature_layer,
        training=False)
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
150
151
152

    embeddings = numpy.zeros(shape=(image.shape[0], embbeding_dim))
    for i in range(image.shape[0]):
153
154
155
156
157
        feed_dict = {
            architecture.inference_placeholder: image[i:i + 1, :, :, :]}
        embedding = session.run(
            [tf.nn.l2_normalize(inference_graph, 1, 1e-10)],
            feed_dict=feed_dict)[0]
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
158
159
160
161
        embedding = numpy.reshape(embedding, numpy.prod(embedding.shape[1:]))
        embeddings[i] = embedding

    return embeddings
162
163
164


def cdist(A):
165
166
167
168
    """
    Compute a pairwise euclidean distance in the same fashion
    as in scipy.spation.distance.cdist
    """
169
    with tf.variable_scope('Pairwisedistance'):
170
171
        ones_1 = tf.reshape(
            tf.cast(tf.ones_like(A), tf.float32)[:, 0], [1, -1])
172
173
        p1 = tf.matmul(
            tf.expand_dims(tf.reduce_sum(tf.square(A), 1), 1),
174
            ones_1
175
        )
176

177
178
        ones_2 = tf.reshape(
            tf.cast(tf.ones_like(A), tf.float32)[:, 0], [-1, 1])
179
180
        p2 = tf.transpose(tf.matmul(
            tf.reshape(tf.reduce_sum(tf.square(A), 1), shape=[-1, 1]),
181
            ones_2,
182
183
184
185
186
187
            transpose_b=True
        ))

        return tf.sqrt(tf.add(p1, p2) - 2 * tf.matmul(A, A, transpose_b=True))


188
def predict_using_tensors(embedding, labels, num=None):
189
    """
190
191
    Compute the predictions through exhaustive comparisons between
    embeddings using tensors
192
193
    """

194
195
    # Fitting the main diagonal with infs (removing comparisons with the same
    # sample)
196
    inf = tf.cast(tf.ones_like(labels), tf.float32) * numpy.inf
197
198
199
200

    distances = cdist(embedding)
    distances = tf.matrix_set_diag(distances, inf)
    indexes = tf.argmin(distances, axis=1)
201
202
203
204
205
    return [labels[i] for i in tf.unstack(indexes, num=num)]


def compute_embedding_accuracy_tensors(embedding, labels, num=None):
    """
206
207
    Compute the accuracy through exhaustive comparisons between the embeddings
    using tensors
208
209
    """

210
211
    # Fitting the main diagonal with infs (removing comparisons with the same
    # sample)
212
    predictions = predict_using_tensors(embedding, labels, num=num)
213
214
    matching = [tf.equal(p, l) for p, l in zip(tf.unstack(
        predictions, num=num), tf.unstack(labels, num=num))]
215

216
    return tf.reduce_sum(tf.cast(matching, tf.uint8)) / len(predictions)
217
218


219
220
def compute_embedding_accuracy(embedding, labels):
    """
221
    Compute the accuracy through exhaustive comparisons between the embeddings
222
223
224
    """

    from scipy.spatial.distance import cdist
225

226
    distances = cdist(embedding, embedding)
227

228
229
    n_samples = embedding.shape[0]

230
231
    # Fitting the main diagonal with infs (removing comparisons with the same
    # sample)
232
    numpy.fill_diagonal(distances, numpy.inf)
233

234
    indexes = distances.argmin(axis=1)
235

236
237
    # Computing the argmin excluding comparisons with the same samples
    # Basically, we are excluding the main diagonal
238
239

    #valid_indexes = distances[distances>0].reshape(n_samples, n_samples-1).argmin(axis=1)
240
241

    # Getting the original positions of the indexes in the 1-axis
242
    #corrected_indexes = [ i if i<j else i+1 for i, j in zip(valid_indexes, range(n_samples))]
243

244
245
246
247
    matching = [labels[i] == labels[j]
                for i, j in zip(range(n_samples), indexes)]
    accuracy = sum(matching) / float(n_samples)

248
    return accuracy
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
249

250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324

def get_available_gpus():
    """Returns the number of GPU devices that are available.

    Returns
    -------
    [str]
        The names of available GPU devices.
    """
    local_device_protos = device_lib.list_local_devices()
    return [x.name for x in local_device_protos if x.device_type == 'GPU']


def to_channels_last(image):
    """Converts the image to channel_last format. This is the same format as in
    matplotlib, skimage, and etc.

    Parameters
    ----------
    image : :any:`tf.Tensor`
        At least a 3 dimensional image. If the dimension is more than 3, the
        last 3 dimensions are assumed to be [C, H, W].

    Returns
    -------
    image : :any:`tf.Tensor`
        The image in [..., H, W, C] format.

    Raises
    ------
    ValueError
        If dim of image is less than 3.
    """
    ndim = len(image.shape)
    if ndim < 3:
        raise ValueError("The image needs to be at least 3 dimensional but it "
                         "was {}".format(ndim))
    axis_order = [1, 2, 0]
    shift = ndim - 3
    axis_order = list(range(ndim - 3)) + [n + shift for n in axis_order]
    return tf.transpose(image, axis_order)


def to_channels_first(image):
    """Converts the image to channel_first format. This is the same format as
    in bob.io.image and bob.io.video.

    Parameters
    ----------
    image : :any:`tf.Tensor`
        At least a 3 dimensional image. If the dimension is more than 3, the
        last 3 dimensions are assumed to be [H, W, C].

    Returns
    -------
    image : :any:`tf.Tensor`
        The image in [..., C, H, W] format.

    Raises
    ------
    ValueError
        If dim of image is less than 3.
    """
    ndim = len(image.shape)
    if ndim < 3:
        raise ValueError("The image needs to be at least 3 dimensional but it "
                         "was {}".format(ndim))
    axis_order = [2, 0, 1]
    shift = ndim - 3
    axis_order = list(range(ndim - 3)) + [n + shift for n in axis_order]
    return tf.transpose(image, axis_order)


to_skimage = to_matplotlib = to_channels_last
to_bob = to_channels_first