File size: 12,547 Bytes
b36e9ec |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 |
"""
Code from https://github.com/hassony2/torch_videovision
"""
import numbers
import random
import numpy as np
import PIL
from skimage.transform import resize, rotate
from skimage.util import pad
import torchvision
import warnings
from skimage import img_as_ubyte, img_as_float
def crop_clip(clip, min_h, min_w, h, w):
if isinstance(clip[0], np.ndarray):
cropped = [img[min_h:min_h + h, min_w:min_w + w, :] for img in clip]
elif isinstance(clip[0], PIL.Image.Image):
cropped = [
img.crop((min_w, min_h, min_w + w, min_h + h)) for img in clip
]
else:
raise TypeError('Expected numpy.ndarray or PIL.Image' +
'but got list of {0}'.format(type(clip[0])))
return cropped
def pad_clip(clip, h, w):
im_h, im_w = clip[0].shape[:2]
pad_h = (0, 0) if h < im_h else ((h - im_h) // 2, (h - im_h + 1) // 2)
pad_w = (0, 0) if w < im_w else ((w - im_w) // 2, (w - im_w + 1) // 2)
return pad(clip, ((0, 0), pad_h, pad_w, (0, 0)), mode='edge')
def resize_clip(clip, size, interpolation='bilinear'):
if isinstance(clip[0], np.ndarray):
if isinstance(size, numbers.Number):
im_h, im_w, im_c = clip[0].shape
# Min spatial dim already matches minimal size
if (im_w <= im_h and im_w == size) or (im_h <= im_w
and im_h == size):
return clip
new_h, new_w = get_resize_sizes(im_h, im_w, size)
size = (new_w, new_h)
else:
size = size[1], size[0]
scaled = [
resize(img, size, order=1 if interpolation == 'bilinear' else 0, preserve_range=True,
mode='constant', anti_aliasing=True) for img in clip
]
elif isinstance(clip[0], PIL.Image.Image):
if isinstance(size, numbers.Number):
im_w, im_h = clip[0].size
# Min spatial dim already matches minimal size
if (im_w <= im_h and im_w == size) or (im_h <= im_w
and im_h == size):
return clip
new_h, new_w = get_resize_sizes(im_h, im_w, size)
size = (new_w, new_h)
else:
size = size[1], size[0]
if interpolation == 'bilinear':
pil_inter = PIL.Image.NEAREST
else:
pil_inter = PIL.Image.BILINEAR
scaled = [img.resize(size, pil_inter) for img in clip]
else:
raise TypeError('Expected numpy.ndarray or PIL.Image' +
'but got list of {0}'.format(type(clip[0])))
return scaled
def get_resize_sizes(im_h, im_w, size):
if im_w < im_h:
ow = size
oh = int(size * im_h / im_w)
else:
oh = size
ow = int(size * im_w / im_h)
return oh, ow
class RandomFlip(object):
def __init__(self, time_flip=False, horizontal_flip=False):
self.time_flip = time_flip
self.horizontal_flip = horizontal_flip
def __call__(self, clip):
if random.random() < 0.5 and self.time_flip:
return clip[::-1]
if random.random() < 0.5 and self.horizontal_flip:
return [np.fliplr(img) for img in clip]
return clip
class RandomResize(object):
"""Resizes a list of (H x W x C) numpy.ndarray to the final size
The larger the original image is, the more times it takes to
interpolate
Args:
interpolation (str): Can be one of 'nearest', 'bilinear'
defaults to nearest
size (tuple): (widht, height)
"""
def __init__(self, ratio=(3. / 4., 4. / 3.), interpolation='nearest'):
self.ratio = ratio
self.interpolation = interpolation
def __call__(self, clip):
scaling_factor = random.uniform(self.ratio[0], self.ratio[1])
if isinstance(clip[0], np.ndarray):
im_h, im_w, im_c = clip[0].shape
elif isinstance(clip[0], PIL.Image.Image):
im_w, im_h = clip[0].size
new_w = int(im_w * scaling_factor)
new_h = int(im_h * scaling_factor)
new_size = (new_w, new_h)
resized = resize_clip(
clip, new_size, interpolation=self.interpolation)
return resized
class RandomCrop(object):
"""Extract random crop at the same location for a list of videos
Args:
size (sequence or int): Desired output size for the
crop in format (h, w)
"""
def __init__(self, size):
if isinstance(size, numbers.Number):
size = (size, size)
self.size = size
def __call__(self, clip):
"""
Args:
img (PIL.Image or numpy.ndarray): List of videos to be cropped
in format (h, w, c) in numpy.ndarray
Returns:
PIL.Image or numpy.ndarray: Cropped list of videos
"""
h, w = self.size
if isinstance(clip[0], np.ndarray):
im_h, im_w, im_c = clip[0].shape
elif isinstance(clip[0], PIL.Image.Image):
im_w, im_h = clip[0].size
else:
raise TypeError('Expected numpy.ndarray or PIL.Image' +
'but got list of {0}'.format(type(clip[0])))
clip = pad_clip(clip, h, w)
im_h, im_w = clip.shape[1:3]
x1 = 0 if h == im_h else random.randint(0, im_w - w)
y1 = 0 if w == im_w else random.randint(0, im_h - h)
cropped = crop_clip(clip, y1, x1, h, w)
return cropped
class RandomRotation(object):
"""Rotate entire clip randomly by a random angle within
given bounds
Args:
degrees (sequence or int): Range of degrees to select from
If degrees is a number instead of sequence like (min, max),
the range of degrees, will be (-degrees, +degrees).
"""
def __init__(self, degrees):
if isinstance(degrees, numbers.Number):
if degrees < 0:
raise ValueError('If degrees is a single number,'
'must be positive')
degrees = (-degrees, degrees)
else:
if len(degrees) != 2:
raise ValueError('If degrees is a sequence,'
'it must be of len 2.')
self.degrees = degrees
def __call__(self, clip):
"""
Args:
img (PIL.Image or numpy.ndarray): List of videos to be cropped
in format (h, w, c) in numpy.ndarray
Returns:
PIL.Image or numpy.ndarray: Cropped list of videos
"""
angle = random.uniform(self.degrees[0], self.degrees[1])
if isinstance(clip[0], np.ndarray):
rotated = [rotate(image=img, angle=angle, preserve_range=True) for img in clip]
elif isinstance(clip[0], PIL.Image.Image):
rotated = [img.rotate(angle) for img in clip]
else:
raise TypeError('Expected numpy.ndarray or PIL.Image' +
'but got list of {0}'.format(type(clip[0])))
return rotated
class ColorJitter(object):
"""Randomly change the brightness, contrast and saturation and hue of the clip
Args:
brightness (float): How much to jitter brightness. brightness_factor
is chosen uniformly from [max(0, 1 - brightness), 1 + brightness].
contrast (float): How much to jitter contrast. contrast_factor
is chosen uniformly from [max(0, 1 - contrast), 1 + contrast].
saturation (float): How much to jitter saturation. saturation_factor
is chosen uniformly from [max(0, 1 - saturation), 1 + saturation].
hue(float): How much to jitter hue. hue_factor is chosen uniformly from
[-hue, hue]. Should be >=0 and <= 0.5.
"""
def __init__(self, brightness=0, contrast=0, saturation=0, hue=0):
self.brightness = brightness
self.contrast = contrast
self.saturation = saturation
self.hue = hue
def get_params(self, brightness, contrast, saturation, hue):
if brightness > 0:
brightness_factor = random.uniform(
max(0, 1 - brightness), 1 + brightness)
else:
brightness_factor = None
if contrast > 0:
contrast_factor = random.uniform(
max(0, 1 - contrast), 1 + contrast)
else:
contrast_factor = None
if saturation > 0:
saturation_factor = random.uniform(
max(0, 1 - saturation), 1 + saturation)
else:
saturation_factor = None
if hue > 0:
hue_factor = random.uniform(-hue, hue)
else:
hue_factor = None
return brightness_factor, contrast_factor, saturation_factor, hue_factor
def __call__(self, clip):
"""
Args:
clip (list): list of PIL.Image
Returns:
list PIL.Image : list of transformed PIL.Image
"""
if isinstance(clip[0], np.ndarray):
brightness, contrast, saturation, hue = self.get_params(
self.brightness, self.contrast, self.saturation, self.hue)
# Create img transform function sequence
img_transforms = []
if brightness is not None:
img_transforms.append(lambda img: torchvision.transforms.functional.adjust_brightness(img, brightness))
if saturation is not None:
img_transforms.append(lambda img: torchvision.transforms.functional.adjust_saturation(img, saturation))
if hue is not None:
img_transforms.append(lambda img: torchvision.transforms.functional.adjust_hue(img, hue))
if contrast is not None:
img_transforms.append(lambda img: torchvision.transforms.functional.adjust_contrast(img, contrast))
random.shuffle(img_transforms)
img_transforms = [img_as_ubyte, torchvision.transforms.ToPILImage()] + img_transforms + [np.array,
img_as_float]
with warnings.catch_warnings():
warnings.simplefilter("ignore")
jittered_clip = []
for img in clip:
jittered_img = img
for func in img_transforms:
jittered_img = func(jittered_img)
jittered_clip.append(jittered_img.astype('float32'))
elif isinstance(clip[0], PIL.Image.Image):
brightness, contrast, saturation, hue = self.get_params(
self.brightness, self.contrast, self.saturation, self.hue)
# Create img transform function sequence
img_transforms = []
if brightness is not None:
img_transforms.append(lambda img: torchvision.transforms.functional.adjust_brightness(img, brightness))
if saturation is not None:
img_transforms.append(lambda img: torchvision.transforms.functional.adjust_saturation(img, saturation))
if hue is not None:
img_transforms.append(lambda img: torchvision.transforms.functional.adjust_hue(img, hue))
if contrast is not None:
img_transforms.append(lambda img: torchvision.transforms.functional.adjust_contrast(img, contrast))
random.shuffle(img_transforms)
# Apply to all videos
jittered_clip = []
for img in clip:
for func in img_transforms:
jittered_img = func(img)
jittered_clip.append(jittered_img)
else:
raise TypeError('Expected numpy.ndarray or PIL.Image' +
'but got list of {0}'.format(type(clip[0])))
return jittered_clip
class AllAugmentationTransform:
def __init__(self, resize_param=None, rotation_param=None, flip_param=None, crop_param=None, jitter_param=None):
self.transforms = []
if flip_param is not None:
self.transforms.append(RandomFlip(**flip_param))
if rotation_param is not None:
self.transforms.append(RandomRotation(**rotation_param))
if resize_param is not None:
self.transforms.append(RandomResize(**resize_param))
if crop_param is not None:
self.transforms.append(RandomCrop(**crop_param))
if jitter_param is not None:
self.transforms.append(ColorJitter(**jitter_param))
def __call__(self, clip):
for t in self.transforms:
clip = t(clip)
return clip
|