Spaces:
Runtime error
Runtime error
File size: 9,970 Bytes
153628e |
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 |
# Copyright (C) 2021-2024, Mindee.
# This program is licensed under the Apache License 2.0.
# See LICENSE or go to <https://opensource.org/licenses/Apache-2.0> for full license details.
import math
import random
from typing import Any, Callable, List, Optional, Tuple, Union
import numpy as np
from doctr.utils.repr import NestedObject
from .. import functional as F
__all__ = ["SampleCompose", "ImageTransform", "ColorInversion", "OneOf", "RandomApply", "RandomRotate", "RandomCrop"]
class SampleCompose(NestedObject):
"""Implements a wrapper that will apply transformations sequentially on both image and target
.. tabs::
.. tab:: TensorFlow
.. code:: python
>>> import numpy as np
>>> import tensorflow as tf
>>> from doctr.transforms import SampleCompose, ImageTransform, ColorInversion, RandomRotate
>>> transfo = SampleCompose([ImageTransform(ColorInversion((32, 32))), RandomRotate(30)])
>>> out, out_boxes = transfo(tf.random.uniform(shape=[64, 64, 3], minval=0, maxval=1), np.zeros((2, 4)))
.. tab:: PyTorch
.. code:: python
>>> import numpy as np
>>> import torch
>>> from doctr.transforms import SampleCompose, ImageTransform, ColorInversion, RandomRotate
>>> transfos = SampleCompose([ImageTransform(ColorInversion((32, 32))), RandomRotate(30)])
>>> out, out_boxes = transfos(torch.rand(8, 64, 64, 3), np.zeros((2, 4)))
Args:
----
transforms: list of transformation modules
"""
_children_names: List[str] = ["sample_transforms"]
def __init__(self, transforms: List[Callable[[Any, Any], Tuple[Any, Any]]]) -> None:
self.sample_transforms = transforms
def __call__(self, x: Any, target: Any) -> Tuple[Any, Any]:
for t in self.sample_transforms:
x, target = t(x, target)
return x, target
class ImageTransform(NestedObject):
"""Implements a transform wrapper to turn an image-only transformation into an image+target transform
.. tabs::
.. tab:: TensorFlow
.. code:: python
>>> import tensorflow as tf
>>> from doctr.transforms import ImageTransform, ColorInversion
>>> transfo = ImageTransform(ColorInversion((32, 32)))
>>> out, _ = transfo(tf.random.uniform(shape=[64, 64, 3], minval=0, maxval=1), None)
.. tab:: PyTorch
.. code:: python
>>> import torch
>>> from doctr.transforms import ImageTransform, ColorInversion
>>> transfo = ImageTransform(ColorInversion((32, 32)))
>>> out, _ = transfo(torch.rand(8, 64, 64, 3), None)
Args:
----
transform: the image transformation module to wrap
"""
_children_names: List[str] = ["img_transform"]
def __init__(self, transform: Callable[[Any], Any]) -> None:
self.img_transform = transform
def __call__(self, img: Any, target: Any) -> Tuple[Any, Any]:
img = self.img_transform(img)
return img, target
class ColorInversion(NestedObject):
"""Applies the following tranformation to a tensor (image or batch of images):
convert to grayscale, colorize (shift 0-values randomly), and then invert colors
.. tabs::
.. tab:: TensorFlow
.. code:: python
>>> import tensorflow as tf
>>> from doctr.transforms import ColorInversion
>>> transfo = ColorInversion(min_val=0.6)
>>> out = transfo(tf.random.uniform(shape=[8, 64, 64, 3], minval=0, maxval=1))
.. tab:: PyTorch
.. code:: python
>>> import torch
>>> from doctr.transforms import ColorInversion
>>> transfo = ColorInversion(min_val=0.6)
>>> out = transfo(torch.rand(8, 64, 64, 3))
Args:
----
min_val: range [min_val, 1] to colorize RGB pixels
"""
def __init__(self, min_val: float = 0.5) -> None:
self.min_val = min_val
def extra_repr(self) -> str:
return f"min_val={self.min_val}"
def __call__(self, img: Any) -> Any:
return F.invert_colors(img, self.min_val)
class OneOf(NestedObject):
"""Randomly apply one of the input transformations
.. tabs::
.. tab:: TensorFlow
.. code:: python
>>> import tensorflow as tf
>>> from doctr.transforms import OneOf
>>> transfo = OneOf([JpegQuality(), Gamma()])
>>> out = transfo(tf.random.uniform(shape=[64, 64, 3], minval=0, maxval=1))
.. tab:: PyTorch
.. code:: python
>>> import torch
>>> from doctr.transforms import OneOf
>>> transfo = OneOf([JpegQuality(), Gamma()])
>>> out = transfo(torch.rand(1, 64, 64, 3))
Args:
----
transforms: list of transformations, one only will be picked
"""
_children_names: List[str] = ["transforms"]
def __init__(self, transforms: List[Callable[[Any], Any]]) -> None:
self.transforms = transforms
def __call__(self, img: Any, target: Optional[np.ndarray] = None) -> Union[Any, Tuple[Any, np.ndarray]]:
# Pick transformation
transfo = self.transforms[int(random.random() * len(self.transforms))]
# Apply
return transfo(img) if target is None else transfo(img, target) # type: ignore[call-arg]
class RandomApply(NestedObject):
"""Apply with a probability p the input transformation
.. tabs::
.. tab:: TensorFlow
.. code:: python
>>> import tensorflow as tf
>>> from doctr.transforms import RandomApply
>>> transfo = RandomApply(Gamma(), p=.5)
>>> out = transfo(tf.random.uniform(shape=[64, 64, 3], minval=0, maxval=1))
.. tab:: PyTorch
.. code:: python
>>> import torch
>>> from doctr.transforms import RandomApply
>>> transfo = RandomApply(Gamma(), p=.5)
>>> out = transfo(torch.rand(1, 64, 64, 3))
Args:
----
transform: transformation to apply
p: probability to apply
"""
def __init__(self, transform: Callable[[Any], Any], p: float = 0.5) -> None:
self.transform = transform
self.p = p
def extra_repr(self) -> str:
return f"transform={self.transform}, p={self.p}"
def __call__(self, img: Any, target: Optional[np.ndarray] = None) -> Union[Any, Tuple[Any, np.ndarray]]:
if random.random() < self.p:
return self.transform(img) if target is None else self.transform(img, target) # type: ignore[call-arg]
return img if target is None else (img, target)
class RandomRotate(NestedObject):
"""Randomly rotate a tensor image and its boxes
.. image:: https://doctr-static.mindee.com/models?id=v0.4.0/rotation_illustration.png&src=0
:align: center
Args:
----
max_angle: maximum angle for rotation, in degrees. Angles will be uniformly picked in
[-max_angle, max_angle]
expand: whether the image should be padded before the rotation
"""
def __init__(self, max_angle: float = 5.0, expand: bool = False) -> None:
self.max_angle = max_angle
self.expand = expand
def extra_repr(self) -> str:
return f"max_angle={self.max_angle}, expand={self.expand}"
def __call__(self, img: Any, target: np.ndarray) -> Tuple[Any, np.ndarray]:
angle = random.uniform(-self.max_angle, self.max_angle)
r_img, r_polys = F.rotate_sample(img, target, angle, self.expand)
# Removes deleted boxes
is_kept = (r_polys.max(1) > r_polys.min(1)).sum(1) == 2
return r_img, r_polys[is_kept]
class RandomCrop(NestedObject):
"""Randomly crop a tensor image and its boxes
Args:
----
scale: tuple of floats, relative (min_area, max_area) of the crop
ratio: tuple of float, relative (min_ratio, max_ratio) where ratio = h/w
"""
def __init__(self, scale: Tuple[float, float] = (0.08, 1.0), ratio: Tuple[float, float] = (0.75, 1.33)) -> None:
self.scale = scale
self.ratio = ratio
def extra_repr(self) -> str:
return f"scale={self.scale}, ratio={self.ratio}"
def __call__(self, img: Any, target: np.ndarray) -> Tuple[Any, np.ndarray]:
scale = random.uniform(self.scale[0], self.scale[1])
ratio = random.uniform(self.ratio[0], self.ratio[1])
height, width = img.shape[:2]
# Calculate crop size
crop_area = scale * width * height
aspect_ratio = ratio * (width / height)
crop_width = int(round(math.sqrt(crop_area * aspect_ratio)))
crop_height = int(round(math.sqrt(crop_area / aspect_ratio)))
# Ensure crop size does not exceed image dimensions
crop_width = min(crop_width, width)
crop_height = min(crop_height, height)
# Randomly select crop position
x = random.randint(0, width - crop_width)
y = random.randint(0, height - crop_height)
# relative crop box
crop_box = (x / width, y / height, (x + crop_width) / width, (y + crop_height) / height)
if target.shape[1:] == (4, 2):
min_xy = np.min(target, axis=1)
max_xy = np.max(target, axis=1)
_target = np.concatenate((min_xy, max_xy), axis=1)
else:
_target = target
# Crop image and targets
croped_img, crop_boxes = F.crop_detection(img, _target, crop_box)
# hard fallback if no box is kept
if crop_boxes.shape[0] == 0:
return img, target
# clip boxes
return croped_img, np.clip(crop_boxes, 0, 1)
|