Spaces:
Running
on
Zero
Running
on
Zero
File size: 12,328 Bytes
9b33fca |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 |
"""Photometric transforms."""
from __future__ import annotations
from collections.abc import Callable
import numpy as np
import torch
import torchvision.transforms.v2.functional as TF
from torch import Tensor
from vis4d.common.imports import OPENCV_AVAILABLE
from vis4d.common.typing import NDArrayF32
from vis4d.data.const import CommonKeys as K
from .base import Transform
if OPENCV_AVAILABLE:
import cv2
else:
raise ImportError("cv2 is not installed.")
@Transform(K.images, K.images)
class RandomGamma:
"""Apply Gamma transformation to images.
Args:
gamma_range (tuple[float, float]): Range of gamma values.
image_channel_mode (str, optional): Image channel mode. Defaults to
"RGB".
"""
def __init__(
self,
gamma_range: tuple[float, float] = (1.0, 1.0),
image_channel_mode: str = "RGB",
) -> None:
"""Init function for Gamma."""
self.gamma_range = gamma_range
self.image_channel_mode = image_channel_mode
assert image_channel_mode in {"RGB", "BGR"}, (
"image_channel_mode should be 'RGB' or 'BGR', "
f"got {image_channel_mode}."
)
def __call__(self, images: list[NDArrayF32]) -> list[NDArrayF32]:
"""Call function for Gamma transformation."""
factor = np.random.uniform(self.gamma_range[0], self.gamma_range[1])
return _adjust_images(
images, TF.adjust_gamma, factor, self.image_channel_mode
)
@Transform(K.images, K.images)
class RandomBrightness:
"""Apply Brightness transformation to images.
Args:
brightness_range (tuple[float, float]): Range of brightness values.
image_channel_mode (str, optional): Image channel mode. Defaults to
"RGB".
"""
def __init__(
self,
brightness_range: tuple[float, float] = (1.0, 1.0),
image_channel_mode: str = "RGB",
) -> None:
"""Init function for Brightness."""
self.brightness_range = brightness_range
self.image_channel_mode = image_channel_mode
assert image_channel_mode in {"RGB", "BGR"}, (
"image_channel_mode should be 'RGB' or 'BGR', "
f"got {image_channel_mode}."
)
def __call__(self, images: list[NDArrayF32]) -> list[NDArrayF32]:
"""Call function for Brightness transformation."""
factor = np.random.uniform(
self.brightness_range[0], self.brightness_range[1]
)
return _adjust_images(
images, TF.adjust_brightness, factor, self.image_channel_mode
)
@Transform(K.images, K.images)
class RandomContrast:
"""Apply Contrast transformation to images.
Args:
contrast_range (tuple[float, float]): Range of contrast values.
image_channel_mode (str, optional): Image channel mode. Defaults to
"RGB".
"""
def __init__(
self,
contrast_range: tuple[float, float] = (1.0, 1.0),
image_channel_mode: str = "RGB",
):
"""Init function for Contrast."""
self.contrast_range = contrast_range
self.image_channel_mode = image_channel_mode
assert image_channel_mode in {"RGB", "BGR"}, (
"image_channel_mode should be 'RGB' or 'BGR', "
f"got {image_channel_mode}."
)
def __call__(self, images: list[NDArrayF32]) -> list[NDArrayF32]:
"""Call function for Contrast transformation."""
factor = np.random.uniform(
self.contrast_range[0], self.contrast_range[1]
)
return _adjust_images(
images, TF.adjust_contrast, factor, self.image_channel_mode
)
@Transform(K.images, K.images)
class RandomSaturation:
"""Apply saturation transformation to images.
Args:
saturation_range (tuple[float, float]): Range of saturation values.
image_channel_mode (str, optional): Image channel mode. Defaults to
"RGB".
"""
def __init__(
self,
saturation_range: tuple[float, float] = (1.0, 1.0),
image_channel_mode: str = "RGB",
):
"""Init function for saturation."""
self.saturation_range = saturation_range
self.image_channel_mode = image_channel_mode
assert image_channel_mode in {"RGB", "BGR"}, (
"image_channel_mode should be 'RGB' or 'BGR', "
f"got {image_channel_mode}."
)
def __call__(self, images: list[NDArrayF32]) -> list[NDArrayF32]:
"""Call function for saturation transformation."""
factor = np.random.uniform(
self.saturation_range[0], self.saturation_range[1]
)
return _adjust_images(
images, TF.adjust_saturation, factor, self.image_channel_mode
)
@Transform(K.images, K.images)
class RandomHue:
"""Apply hue transformation to images.
Args:
hue_range (tuple[float, float]): Range of hue values.
image_channel_mode (str, optional): Image channel mode. Defaults to
"RGB".
"""
def __init__(
self,
hue_range: tuple[float, float] = (0.0, 0.0),
image_channel_mode: str = "RGB",
):
"""Init function for hue."""
self.hue_range = hue_range
self.image_channel_mode = image_channel_mode
assert image_channel_mode in {"RGB", "BGR"}, (
"image_channel_mode should be 'RGB' or 'BGR', "
f"got {image_channel_mode}."
)
def __call__(self, images: list[NDArrayF32]) -> list[NDArrayF32]:
"""Call function for Hue transformation."""
factor = np.random.uniform(self.hue_range[0], self.hue_range[1])
return _adjust_images(
images, TF.adjust_hue, factor, self.image_channel_mode
)
@Transform(K.images, K.images)
class ColorJitter:
"""Apply color jitter to images.
Args:
brightness_range (tuple[float, float]): Range of brightness values.
contrast_range (tuple[float, float]): Range of contrast values.
saturation_range (tuple[float, float]): Range of saturation values.
hue_range (tuple[float, float]): Range of hue values.
image_channel_mode (str, optional): Image channel mode. Defaults to
"RGB".
"""
def __init__(
self,
brightness_range: tuple[float, float] = (0.875, 1.125),
contrast_range: tuple[float, float] = (0.5, 1.5),
saturation_range: tuple[float, float] = (0.5, 1.5),
hue_range: tuple[float, float] = (-0.05, 0.05),
image_channel_mode: str = "RGB",
):
"""Init function for color jitter."""
self.brightness_range = brightness_range
self.contrast_range = contrast_range
self.saturation_range = saturation_range
self.hue_range = hue_range
self.image_channel_mode = image_channel_mode
assert image_channel_mode in {"RGB", "BGR"}, (
"image_channel_mode should be 'RGB' or 'BGR', "
f"got {image_channel_mode}."
)
def __call__(self, images: list[NDArrayF32]) -> list[NDArrayF32]:
"""Call function for Hue transformation."""
transform_order = np.random.permutation(4)
for transform in transform_order:
# apply photometric transforms in a random order
if transform == 0:
# random brightness
bfactor = np.random.uniform(
self.brightness_range[0], self.brightness_range[1]
)
images = _adjust_images(
images,
TF.adjust_brightness,
bfactor,
self.image_channel_mode,
)
elif transform == 1:
# random contrast
cfactor = np.random.uniform(
self.contrast_range[0], self.contrast_range[1]
)
images = _adjust_images(
images,
TF.adjust_contrast,
cfactor,
self.image_channel_mode,
)
elif transform == 2:
# random saturation
sfactor = np.random.uniform(
self.saturation_range[0], self.saturation_range[1]
)
images = _adjust_images(
images,
TF.adjust_saturation,
sfactor,
self.image_channel_mode,
)
elif transform == 3:
# random hue
hfactor = np.random.uniform(
self.hue_range[0], self.hue_range[1]
)
images = _adjust_images(
images, TF.adjust_hue, hfactor, self.image_channel_mode
)
return images
def _adjust_images(
images: list[NDArrayF32],
adjust_func: Callable[[Tensor, float], Tensor],
adj_factor: float,
image_channel_mode: str = "RGB",
) -> list[NDArrayF32]:
"""Apply color transformation to images.
Args:
images (list[NDArrayF32]): Image to be transformed.
adjust_func (Callable[[Tensor, float], Tensor]): Function to apply.
adj_factor (float): Adjustment factor.
image_channel_mode (str, optional): Image channel mode. Defaults to
"RGB".
Returns:
list[NDArrayF32]: Transformed image.
"""
for i, image in enumerate(images):
if image_channel_mode == "BGR":
image = image[..., [2, 1, 0]] # convert to RGB
image_ = torch.from_numpy(image).permute(0, 3, 1, 2) / 255.0
image_ = adjust_func(image_, adj_factor) * 255.0
images[i] = image_.permute(0, 2, 3, 1).numpy()
if image_channel_mode == "BGR":
images[i] = images[i][..., [2, 1, 0]] # convert back to BGR
return images
@Transform(K.images, K.images)
class RandomHSV:
"""Apply HSV transformation to images.
Used by YOLOX. Modifed from: https://github.com/Megvii-BaseDetection/YOLOX.
Args:
hue_delta (int): Delta for hue.
saturation_delta (int): Delta for saturation.
value_delta (int): Delta for value.
image_channel_mode (str, optional): Image channel mode. Defaults to
"BGR".
"""
def __init__(
self,
hue_delta: int = 5,
saturation_delta: int = 30,
value_delta: int = 30,
image_channel_mode: str = "BGR",
):
"""Init function for HSV transformation."""
assert OPENCV_AVAILABLE, "RandomHSV requires OpenCV to be installed."
self.hue_delta = hue_delta
self.saturation_delta = saturation_delta
self.value_delta = value_delta
self.image_channel_mode = image_channel_mode
assert image_channel_mode in {"RGB", "BGR"}, (
"image_channel_mode should be 'RGB' or 'BGR', "
f"got {image_channel_mode}."
)
# pylint: disable=no-member
def __call__(self, images: list[NDArrayF32]) -> list[NDArrayF32]:
"""Call function for Hue transformation."""
for i, image in enumerate(images):
image = image[0].astype(np.uint8)
if self.image_channel_mode == "BGR":
image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
else:
image = cv2.cvtColor(image, cv2.COLOR_RGB2HSV)
image = image.astype(np.int16)
hsv_gains = np.random.uniform(-1, 1, 3) * [
self.hue_delta,
self.saturation_delta,
self.value_delta,
]
# random selection of h, s, v
hsv_gains = (hsv_gains * np.random.randint(0, 2, 3)).astype(
np.int16
)
image[..., 0] = (image[..., 0] + hsv_gains[0]) % 180
image[..., 1] = np.clip(image[..., 1] + hsv_gains[1], 0, 255)
image[..., 2] = np.clip(image[..., 2] + hsv_gains[2], 0, 255)
image = image.astype(np.uint8)
if self.image_channel_mode == "BGR":
cv2.cvtColor(image, cv2.COLOR_HSV2BGR, dst=image)
else:
cv2.cvtColor(image, cv2.COLOR_HSV2RGB, dst=image)
images[i] = image[None, ...].astype(np.float32)
return images
|