This repository was archived by the owner on Mar 10, 2026. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 327
Expand file tree
/
Copy pathcut_mix.py
More file actions
162 lines (138 loc) · 5.85 KB
/
cut_mix.py
File metadata and controls
162 lines (138 loc) · 5.85 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
# Copyright 2022 The KerasCV Authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import tensorflow as tf
from tensorflow import keras
from keras_cv.layers.preprocessing.base_image_augmentation_layer import (
BaseImageAugmentationLayer,
)
from keras_cv.utils import fill_utils
@keras.utils.register_keras_serializable(package="keras_cv")
class CutMix(BaseImageAugmentationLayer):
"""CutMix implements the CutMix data augmentation technique.
Args:
alpha: Float between 0 and 1. Inverse scale parameter for the gamma
distribution. This controls the shape of the distribution from which
the smoothing values are sampled. 1.0 is a recommended value when
training an imagenet1k classification model. Defaults to `1.0`.
seed: Integer. Used to create a random seed.
References:
- [CutMix paper]( https://arxiv.org/abs/1905.04899).
Sample usage:
```python
(images, labels), _ = keras.datasets.cifar10.load_data()
labels = tf.one_hot(labels.squeeze(), 10)
cutmix = keras_cv.layers.preprocessing.cut_mix.CutMix(10)
output = cutmix({"images": images[:32], "labels": labels[:32]})
# output == {'images': updated_images, 'labels': updated_labels}
```
"""
def __init__(self, alpha=1.0, seed=None, **kwargs):
super().__init__(seed=seed, **kwargs)
self.alpha = alpha
self.seed = seed
def _sample_from_beta(self, alpha, beta, shape):
sample_alpha = tf.random.gamma(
shape, alpha=alpha, seed=self._random_generator.make_legacy_seed()
)
sample_beta = tf.random.gamma(
shape, alpha=beta, seed=self._random_generator.make_legacy_seed()
)
return sample_alpha / (sample_alpha + sample_beta)
def _batch_augment(self, inputs):
self._validate_inputs(inputs)
images = inputs.get("images", None)
labels = inputs.get("labels", None)
if images is None or labels is None:
raise ValueError(
"CutMix expects inputs in a dictionary with format "
'{"images": images, "labels": labels}.'
f"Got: inputs = {inputs}"
)
images, labels = self._update_labels(*self._cutmix(images, labels))
inputs["images"] = images
inputs["labels"] = labels
return inputs
def _augment(self, inputs):
raise ValueError(
"CutMix received a single image to `call`. The layer relies on "
"combining multiple examples, and as such will not behave as "
"expected. Please call the layer with 2 or more samples."
)
def _cutmix(self, images, labels):
"""Apply cutmix."""
input_shape = tf.shape(images)
batch_size, image_height, image_width = (
input_shape[0],
input_shape[1],
input_shape[2],
)
permutation_order = tf.random.shuffle(
tf.range(0, batch_size), seed=self.seed
)
lambda_sample = self._sample_from_beta(
self.alpha, self.alpha, (batch_size,)
)
ratio = tf.math.sqrt(1 - lambda_sample)
cut_height = tf.cast(
ratio * tf.cast(image_height, dtype=tf.float32), dtype=tf.int32
)
cut_width = tf.cast(
ratio * tf.cast(image_width, dtype=tf.float32), dtype=tf.int32
)
random_center_height = tf.random.uniform(
shape=[batch_size], minval=0, maxval=image_height, dtype=tf.int32
)
random_center_width = tf.random.uniform(
shape=[batch_size], minval=0, maxval=image_width, dtype=tf.int32
)
bounding_box_area = cut_height * cut_width
lambda_sample = 1.0 - bounding_box_area / (image_height * image_width)
lambda_sample = tf.cast(lambda_sample, dtype=self.compute_dtype)
images = fill_utils.fill_rectangle(
images,
random_center_width,
random_center_height,
cut_width,
cut_height,
tf.gather(images, permutation_order),
)
return images, labels, lambda_sample, permutation_order
def _update_labels(self, images, labels, lambda_sample, permutation_order):
cutout_labels = tf.gather(labels, permutation_order)
lambda_sample = tf.reshape(lambda_sample, [-1, 1])
labels = lambda_sample * labels + (1.0 - lambda_sample) * cutout_labels
return images, labels
def _validate_inputs(self, inputs):
labels = inputs.get("labels", None)
if labels is None:
raise ValueError(
"CutMix expects 'labels' to be present in its inputs. "
"CutMix relies on both images an labels. "
"Please pass a dictionary with keys 'images' "
"containing the image Tensor, and 'labels' containing "
"the classification labels. "
"For example, `cut_mix({'images': images, 'labels': labels})`."
)
if not labels.dtype.is_floating:
raise ValueError(
f"CutMix received labels with type {labels.dtype}. "
"Labels must be of type float."
)
def get_config(self):
config = {
"alpha": self.alpha,
"seed": self.seed,
}
base_config = super().get_config()
return dict(list(base_config.items()) + list(config.items()))