MAERec-Gradio / mmocr /datasets /transforms /textdet_transforms.py
Mountchicken's picture
Upload 704 files
9bf4bd7
raw
history blame
32.1 kB
# Copyright (c) OpenMMLab. All rights reserved.
import math
import random
from typing import Dict, List, Sequence, Tuple, Union
import cv2
import mmcv
import numpy as np
from mmcv.transforms import RandomFlip as MMCV_RandomFlip
from mmcv.transforms.base import BaseTransform
from mmcv.transforms.utils import avoid_cache_randomness, cache_randomness
from shapely.geometry import Polygon as plg
from mmocr.registry import TRANSFORMS
from mmocr.utils import crop_polygon, poly2bbox, poly_intersection
@TRANSFORMS.register_module()
@avoid_cache_randomness
class BoundedScaleAspectJitter(BaseTransform):
"""First randomly rescale the image so that the longside and shortside of
the image are around the bound; then jitter its aspect ratio.
Required Keys:
- img
- img_shape
- gt_bboxes (optional)
- gt_polygons (optional)
Modified Keys:
- img
- img_shape
- gt_bboxes (optional)
- gt_polygons (optional)
Added Keys:
- scale
- scale_factor
- keep_ratio
Args:
long_size_bound (int): The approximate bound for long size.
short_size_bound (int): The approximate bound for short size.
size_jitter_range (tuple(float, float)): Range of the ratio used
to jitter the size. Defaults to (0.7, 1.3).
aspect_ratio_jitter_range (tuple(float, float)): Range of the ratio
used to jitter its aspect ratio. Defaults to (0.9, 1.1).
resize_type (str): The type of resize class to use. Defaults to
"Resize".
**resize_kwargs: Other keyword arguments for the ``resize_type``.
"""
def __init__(
self,
long_size_bound: int,
short_size_bound: int,
ratio_range: Tuple[float, float] = (0.7, 1.3),
aspect_ratio_range: Tuple[float, float] = (0.9, 1.1),
resize_type: str = 'Resize',
**resize_kwargs,
) -> None:
super().__init__()
self.ratio_range = ratio_range
self.aspect_ratio_range = aspect_ratio_range
self.long_size_bound = long_size_bound
self.short_size_bound = short_size_bound
self.resize_cfg = dict(type=resize_type, **resize_kwargs)
# create an empty Reisize object
self.resize_cfg.update(dict(scale=0))
self.resize = TRANSFORMS.build(self.resize_cfg)
def _sample_from_range(self, range: Tuple[float, float]) -> float:
"""A ratio will be randomly sampled from the range specified by
``range``.
Args:
ratio_range (tuple[float]): The minimum and maximum ratio.
Returns:
float: A ratio randomly sampled from the range.
"""
min_value, max_value = min(range), max(range)
value = np.random.random_sample() * (max_value - min_value) + min_value
return value
def transform(self, results: Dict) -> Dict:
h, w = results['img'].shape[:2]
new_scale = 1
if max(h, w) > self.long_size_bound:
new_scale = self.long_size_bound / max(h, w)
jitter_ratio = self._sample_from_range(self.ratio_range)
jitter_ratio = new_scale * jitter_ratio
if min(h, w) * jitter_ratio <= self.short_size_bound:
jitter_ratio = (self.short_size_bound + 10) * 1.0 / min(h, w)
aspect = self._sample_from_range(self.aspect_ratio_range)
h_scale = jitter_ratio * math.sqrt(aspect)
w_scale = jitter_ratio / math.sqrt(aspect)
new_h = int(h * h_scale)
new_w = int(w * w_scale)
self.resize.scale = (new_w, new_h)
return self.resize(results)
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(long_size_bound = {self.long_size_bound}, '
repr_str += f'short_size_bound = {self.short_size_bound}, '
repr_str += f'ratio_range = {self.ratio_range}, '
repr_str += f'aspect_ratio_range = {self.aspect_ratio_range}, '
repr_str += f'resize_cfg = {self.resize_cfg})'
return repr_str
@TRANSFORMS.register_module()
class RandomFlip(MMCV_RandomFlip):
"""Flip the image & bbox polygon.
There are 3 flip modes:
- ``prob`` is float, ``direction`` is string: the image will be
``direction``ly flipped with probability of ``prob`` .
E.g., ``prob=0.5``, ``direction='horizontal'``,
then image will be horizontally flipped with probability of 0.5.
- ``prob`` is float, ``direction`` is list of string: the image will
be ``direction[i]``ly flipped with probability of
``prob/len(direction)``.
E.g., ``prob=0.5``, ``direction=['horizontal', 'vertical']``,
then image will be horizontally flipped with probability of 0.25,
vertically with probability of 0.25.
- ``prob`` is list of float, ``direction`` is list of string:
given ``len(prob) == len(direction)``, the image will
be ``direction[i]``ly flipped with probability of ``prob[i]``.
E.g., ``prob=[0.3, 0.5]``, ``direction=['horizontal',
'vertical']``, then image will be horizontally flipped with
probability of 0.3, vertically with probability of 0.5.
Required Keys:
- img
- gt_bboxes (optional)
- gt_polygons (optional)
Modified Keys:
- img
- gt_bboxes (optional)
- gt_polygons (optional)
Added Keys:
- flip
- flip_direction
Args:
prob (float | list[float], optional): The flipping probability.
Defaults to None.
direction(str | list[str]): The flipping direction. Options
If input is a list, the length must equal ``prob``. Each
element in ``prob`` indicates the flip probability of
corresponding direction. Defaults to 'horizontal'.
"""
def flip_polygons(self, polygons: Sequence[np.ndarray],
img_shape: Tuple[int, int],
direction: str) -> Sequence[np.ndarray]:
"""Flip polygons horizontally, vertically or diagonally.
Args:
polygons (list[numpy.ndarray): polygons.
img_shape (tuple[int]): Image shape (height, width)
direction (str): Flip direction. Options are 'horizontal',
'vertical' and 'diagonal'.
Returns:
list[numpy.ndarray]: Flipped polygons.
"""
h, w = img_shape
flipped_polygons = []
if direction == 'horizontal':
for polygon in polygons:
flipped_polygon = polygon.copy()
flipped_polygon[0::2] = w - polygon[0::2]
flipped_polygons.append(flipped_polygon)
elif direction == 'vertical':
for polygon in polygons:
flipped_polygon = polygon.copy()
flipped_polygon[1::2] = h - polygon[1::2]
flipped_polygons.append(flipped_polygon)
elif direction == 'diagonal':
for polygon in polygons:
flipped_polygon = polygon.copy()
flipped_polygon[0::2] = w - polygon[0::2]
flipped_polygon[1::2] = h - polygon[1::2]
flipped_polygons.append(flipped_polygon)
else:
raise ValueError(
f"Flipping direction must be 'horizontal', 'vertical', \
or 'diagnal', but got '{direction}'")
return flipped_polygons
def _flip(self, results: dict) -> None:
"""Flip images, bounding boxes and polygons.
Args:
results (dict): Result dict containing the data to transform.
"""
super()._flip(results)
# flip polygons
if results.get('gt_polygons', None) is not None:
results['gt_polygons'] = self.flip_polygons(
results['gt_polygons'], results['img'].shape[:2],
results['flip_direction'])
@TRANSFORMS.register_module()
class SourceImagePad(BaseTransform):
"""Pad Image to target size. It will randomly crop an area from the
original image and resize it to the target size, then paste the original
image to its top left corner.
Required Keys:
- img
Modified Keys:
- img
- img_shape
Added Keys:
- pad_shape
- pad_fixed_size
Args:
target_scale (int or tuple[int, int]]): The target size of padded
image. If it's an integer, then the padding size would be
(target_size, target_size). If it's tuple, then ``target_scale[0]``
should be the width and ``target_scale[1]`` should be the height.
The size of the padded image will be (target_scale[1],
target_scale[0])
crop_ratio (float or Tuple[float, float]): Relative size for the
crop region. If ``crop_ratio`` is a float, then the initial crop
size would be
``(crop_ratio * img.shape[0], crop_ratio * img.shape[1])`` . If
``crop_ratio`` is a tuple, then ``crop_ratio[0]`` is for the width
and ``crop_ratio[1]`` is for the height. The initial crop size
would be
``(crop_ratio[1] * img.shape[0], crop_ratio[0] * img.shape[1])``.
Defaults to 1./9.
"""
def __init__(self,
target_scale: Union[int, Tuple[int, int]],
crop_ratio: Union[float, Tuple[float,
float]] = 1. / 9) -> None:
self.target_scale = target_scale if isinstance(
target_scale, tuple) else (target_scale, target_scale)
self.crop_ratio = crop_ratio if isinstance(
crop_ratio, tuple) else (crop_ratio, crop_ratio)
def transform(self, results: Dict) -> Dict:
"""Pad Image to target size. It will randomly select a small area from
the original image and resize it to the target size, then paste the
original image to its top left corner.
Args:
results (Dict): Result dict containing the data to transform.
Returns:
(Dict): The transformed data.
"""
img = results['img']
h, w = img.shape[:2]
assert h <= self.target_scale[1] and w <= self.target_scale[
0], 'image size should be smaller that the target size'
h_ind = np.random.randint(0, int(h - h * self.crop_ratio[1]) + 1)
w_ind = np.random.randint(0, int(w - w * self.crop_ratio[0]) + 1)
img_cut = img[h_ind:int(h_ind + h * self.crop_ratio[1]),
w_ind:int(w_ind + w * self.crop_ratio[1])]
expand_img = mmcv.imresize(img_cut, self.target_scale)
# paste img to the top left corner of the padding region
expand_img[0:h, 0:w] = img
results['img'] = expand_img
results['img_shape'] = expand_img.shape[:2]
results['pad_shape'] = expand_img.shape
results['pad_fixed_size'] = self.target_scale
return results
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(target_scale = {self.target_scale}, '
repr_str += f'crop_ratio = {self.crop_ratio})'
return repr_str
@TRANSFORMS.register_module()
@avoid_cache_randomness
class ShortScaleAspectJitter(BaseTransform):
"""First rescale the image for its shorter side to reach the short_size and
then jitter its aspect ratio, final rescale the shape guaranteed to be
divided by scale_divisor.
Required Keys:
- img
- img_shape
- gt_bboxes (optional)
- gt_polygons (optional)
Modified Keys:
- img
- img_shape
- gt_bboxes (optional)
- gt_polygons (optional)
Added Keys:
- scale
- scale_factor
- keep_ratio
Args:
short_size (int): Target shorter size before jittering the aspect
ratio. Defaults to 736.
short_size_jitter_range (tuple(float, float)): Range of the ratio used
to jitter the target shorter size. Defaults to (0.7, 1.3).
aspect_ratio_jitter_range (tuple(float, float)): Range of the ratio
used to jitter its aspect ratio. Defaults to (0.9, 1.1).
scale_divisor (int): The scale divisor. Defaults to 1.
resize_type (str): The type of resize class to use. Defaults to
"Resize".
**resize_kwargs: Other keyword arguments for the ``resize_type``.
"""
def __init__(self,
short_size: int = 736,
ratio_range: Tuple[float, float] = (0.7, 1.3),
aspect_ratio_range: Tuple[float, float] = (0.9, 1.1),
scale_divisor: int = 1,
resize_type: str = 'Resize',
**resize_kwargs) -> None:
super().__init__()
self.short_size = short_size
self.ratio_range = ratio_range
self.aspect_ratio_range = aspect_ratio_range
self.resize_cfg = dict(type=resize_type, **resize_kwargs)
# create a empty Reisize object
self.resize_cfg.update(dict(scale=0))
self.resize = TRANSFORMS.build(self.resize_cfg)
self.scale_divisor = scale_divisor
def _sample_from_range(self, range: Tuple[float, float]) -> float:
"""A ratio will be randomly sampled from the range specified by
``range``.
Args:
ratio_range (tuple[float]): The minimum and maximum ratio.
Returns:
float: A ratio randomly sampled from the range.
"""
min_value, max_value = min(range), max(range)
value = np.random.random_sample() * (max_value - min_value) + min_value
return value
def transform(self, results: Dict) -> Dict:
"""Short Scale Aspect Jitter.
Args:
results (dict): Result dict containing the data to transform.
Returns:
dict: The transformed data.
"""
h, w = results['img'].shape[:2]
ratio = self._sample_from_range(self.ratio_range)
scale = (ratio * self.short_size) / min(h, w)
aspect = self._sample_from_range(self.aspect_ratio_range)
h_scale = scale * math.sqrt(aspect)
w_scale = scale / math.sqrt(aspect)
new_h = round(h * h_scale)
new_w = round(w * w_scale)
new_h = math.ceil(new_h / self.scale_divisor) * self.scale_divisor
new_w = math.ceil(new_w / self.scale_divisor) * self.scale_divisor
self.resize.scale = (new_w, new_h)
return self.resize(results)
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(short_size = {self.short_size}, '
repr_str += f'ratio_range = {self.ratio_range}, '
repr_str += f'aspect_ratio_range = {self.aspect_ratio_range}, '
repr_str += f'scale_divisor = {self.scale_divisor}, '
repr_str += f'resize_cfg = {self.resize_cfg})'
return repr_str
@TRANSFORMS.register_module()
class TextDetRandomCropFlip(BaseTransform):
# TODO Rename this transformer; Refactor the redundant code.
"""Random crop and flip a patch in the image. Only used in text detection
task.
Required Keys:
- img
- gt_bboxes
- gt_polygons
Modified Keys:
- img
- gt_bboxes
- gt_polygons
Args:
pad_ratio (float): The ratio of padding. Defaults to 0.1.
crop_ratio (float): The ratio of cropping. Defaults to 0.5.
iter_num (int): Number of operations. Defaults to 1.
min_area_ratio (float): Minimal area ratio between cropped patch
and original image. Defaults to 0.2.
epsilon (float): The threshold of polygon IoU between cropped area
and polygon, which is used to avoid cropping text instances.
Defaults to 0.01.
"""
def __init__(self,
pad_ratio: float = 0.1,
crop_ratio: float = 0.5,
iter_num: int = 1,
min_area_ratio: float = 0.2,
epsilon: float = 1e-2) -> None:
if not isinstance(pad_ratio, float):
raise TypeError('`pad_ratio` should be an float, '
f'but got {type(pad_ratio)} instead')
if not isinstance(crop_ratio, float):
raise TypeError('`crop_ratio` should be a float, '
f'but got {type(crop_ratio)} instead')
if not isinstance(iter_num, int):
raise TypeError('`iter_num` should be an integer, '
f'but got {type(iter_num)} instead')
if not isinstance(min_area_ratio, float):
raise TypeError('`min_area_ratio` should be a float, '
f'but got {type(min_area_ratio)} instead')
if not isinstance(epsilon, float):
raise TypeError('`epsilon` should be a float, '
f'but got {type(epsilon)} instead')
self.pad_ratio = pad_ratio
self.epsilon = epsilon
self.crop_ratio = crop_ratio
self.iter_num = iter_num
self.min_area_ratio = min_area_ratio
@cache_randomness
def _random_prob(self) -> float:
"""Get the random prob to decide whether apply the transform.
Returns:
float: The probability
"""
return random.random()
@cache_randomness
def _random_flip_type(self) -> int:
"""Get the random flip type.
Returns:
int: The flip type index. (0: horizontal; 1: vertical; 2: both)
"""
return np.random.randint(3)
@cache_randomness
def _random_choice(self, axis: np.ndarray) -> np.ndarray:
"""Randomly select two coordinates from the axis.
Args:
axis (np.ndarray): Result dict containing the data to transform
Returns:
np.ndarray: The selected coordinates
"""
return np.random.choice(axis, size=2)
def transform(self, results: Dict) -> Dict:
"""Applying random crop flip on results.
Args:
results (dict): Result dict containing the data to transform
Returns:
dict: The transformed data
"""
assert 'img' in results, '`img` is not found in results'
for _ in range(self.iter_num):
results = self._random_crop_flip_polygons(results)
bboxes = [poly2bbox(poly) for poly in results['gt_polygons']]
results['gt_bboxes'] = np.array(
bboxes, dtype=np.float32).reshape(-1, 4)
return results
def _random_crop_flip_polygons(self, results: Dict) -> Dict:
"""Applying random crop flip on polygons.
Args:
results (dict): Result dict containing the data to transform
Returns:
dict: The transformed data
"""
if results.get('gt_polygons', None) is None:
return results
image = results['img']
polygons = results['gt_polygons']
if len(polygons) == 0 or self._random_prob() > self.crop_ratio:
return results
h, w = results['img_shape']
area = h * w
pad_h = int(h * self.pad_ratio)
pad_w = int(w * self.pad_ratio)
h_axis, w_axis = self._generate_crop_target(image, polygons, pad_h,
pad_w)
if len(h_axis) == 0 or len(w_axis) == 0:
return results
# At most 10 attempts
for _ in range(10):
polys_keep = []
polys_new = []
kept_idxs = []
xx = self._random_choice(w_axis)
yy = self._random_choice(h_axis)
xmin = np.clip(np.min(xx) - pad_w, 0, w - 1)
xmax = np.clip(np.max(xx) - pad_w, 0, w - 1)
ymin = np.clip(np.min(yy) - pad_h, 0, h - 1)
ymax = np.clip(np.max(yy) - pad_h, 0, h - 1)
if (xmax - xmin) * (ymax - ymin) < area * self.min_area_ratio:
# Skip when cropped area is too small
continue
pts = np.stack([[xmin, xmax, xmax, xmin],
[ymin, ymin, ymax, ymax]]).T.astype(np.int32)
pp = plg(pts)
success_flag = True
for poly_idx, polygon in enumerate(polygons):
ppi = plg(polygon.reshape(-1, 2))
ppiou = poly_intersection(ppi, pp)
if np.abs(ppiou - float(ppi.area)) > self.epsilon and \
np.abs(ppiou) > self.epsilon:
success_flag = False
break
kept_idxs.append(poly_idx)
if np.abs(ppiou - float(ppi.area)) < self.epsilon:
polys_new.append(polygon)
else:
polys_keep.append(polygon)
if success_flag:
break
cropped = image[ymin:ymax, xmin:xmax, :]
select_type = self._random_flip_type()
if select_type == 0:
img = np.ascontiguousarray(cropped[:, ::-1])
elif select_type == 1:
img = np.ascontiguousarray(cropped[::-1, :])
else:
img = np.ascontiguousarray(cropped[::-1, ::-1])
image[ymin:ymax, xmin:xmax, :] = img
results['img'] = image
if len(polys_new) != 0:
height, width, _ = cropped.shape
if select_type == 0:
for idx, polygon in enumerate(polys_new):
poly = polygon.reshape(-1, 2)
poly[:, 0] = width - poly[:, 0] + 2 * xmin
polys_new[idx] = poly.reshape(-1, )
elif select_type == 1:
for idx, polygon in enumerate(polys_new):
poly = polygon.reshape(-1, 2)
poly[:, 1] = height - poly[:, 1] + 2 * ymin
polys_new[idx] = poly.reshape(-1, )
else:
for idx, polygon in enumerate(polys_new):
poly = polygon.reshape(-1, 2)
poly[:, 0] = width - poly[:, 0] + 2 * xmin
poly[:, 1] = height - poly[:, 1] + 2 * ymin
polys_new[idx] = poly.reshape(-1, )
polygons = polys_keep + polys_new
# ignored = polys_keep_ignore_idx + polys_new_ignore_idx
results['gt_polygons'] = polygons
results['gt_ignored'] = results['gt_ignored'][kept_idxs]
results['gt_bboxes_labels'] = results['gt_bboxes_labels'][
kept_idxs]
return results
def _generate_crop_target(self, image: np.ndarray,
all_polys: List[np.ndarray], pad_h: int,
pad_w: int) -> Tuple[np.ndarray, np.ndarray]:
"""Generate cropping target and make sure not to crop the polygon
instances.
Args:
image (np.ndarray): The image waited to be crop.
all_polys (list[np.ndarray]): Ground-truth polygons.
pad_h (int): Padding length of height.
pad_w (int): Padding length of width.
Returns:
(np.ndarray, np.ndarray): Returns a tuple ``(h_axis, w_axis)``,
where ``h_axis`` is the vertical cropping range and ``w_axis``
is the horizontal cropping range.
"""
h, w, _ = image.shape
h_array = np.zeros((h + pad_h * 2), dtype=np.int32)
w_array = np.zeros((w + pad_w * 2), dtype=np.int32)
text_polys = []
for polygon in all_polys:
rect = cv2.minAreaRect(polygon.astype(np.int32).reshape(-1, 2))
box = cv2.boxPoints(rect)
box = np.int0(box)
text_polys.append([box[0], box[1], box[2], box[3]])
polys = np.array(text_polys, dtype=np.int32)
for poly in polys:
poly = np.round(poly, decimals=0).astype(np.int32)
minx, maxx = np.min(poly[:, 0]), np.max(poly[:, 0])
miny, maxy = np.min(poly[:, 1]), np.max(poly[:, 1])
w_array[minx + pad_w:maxx + pad_w] = 1
h_array[miny + pad_h:maxy + pad_h] = 1
h_axis = np.where(h_array == 0)[0]
w_axis = np.where(w_array == 0)[0]
return h_axis, w_axis
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(pad_ratio = {self.pad_ratio}'
repr_str += f', crop_ratio = {self.crop_ratio}'
repr_str += f', iter_num = {self.iter_num}'
repr_str += f', min_area_ratio = {self.min_area_ratio}'
repr_str += f', epsilon = {self.epsilon})'
return repr_str
@TRANSFORMS.register_module()
@avoid_cache_randomness
class TextDetRandomCrop(BaseTransform):
"""Randomly select a region and crop images to a target size and make sure
to contain text region. This transform may break up text instances, and for
broken text instances, we will crop it's bbox and polygon coordinates. This
transform is recommend to be used in segmentation-based network.
Required Keys:
- img
- gt_polygons
- gt_bboxes
- gt_bboxes_labels
- gt_ignored
Modified Keys:
- img
- img_shape
- gt_polygons
- gt_bboxes
- gt_bboxes_labels
- gt_ignored
Args:
target_size (tuple(int, int) or int): Target size for the cropped
image. If it's a tuple, then target width and target height will be
``target_size[0]`` and ``target_size[1]``, respectively. If it's an
integer, them both target width and target height will be
``target_size``.
positive_sample_ratio (float): The probability of sampling regions
that go through text regions. Defaults to 5. / 8.
"""
def __init__(self,
target_size: Tuple[int, int] or int,
positive_sample_ratio: float = 5.0 / 8.0) -> None:
self.target_size = target_size if isinstance(
target_size, tuple) else (target_size, target_size)
self.positive_sample_ratio = positive_sample_ratio
def _get_postive_prob(self) -> float:
"""Get the probability to do positive sample.
Returns:
float: The probability to do positive sample.
"""
return np.random.random_sample()
def _sample_num(self, start, end):
"""Sample a number in range [start, end].
Args:
start (int): Starting point.
end (int): Ending point.
Returns:
(int): Sampled number.
"""
return random.randint(start, end)
def _sample_offset(self, gt_polygons: Sequence[np.ndarray],
img_size: Tuple[int, int]) -> Tuple[int, int]:
"""Samples the top-left coordinate of a crop region, ensuring that the
cropped region contains at least one polygon.
Args:
gt_polygons (list(ndarray)) : Polygons.
img_size (tuple(int, int)) : Image size in the format of
(height, width).
Returns:
tuple(int, int): Top-left coordinate of the cropped region.
"""
h, w = img_size
t_w, t_h = self.target_size
# target size is bigger than origin size
t_h = t_h if t_h < h else h
t_w = t_w if t_w < w else w
if (gt_polygons is not None and len(gt_polygons) > 0
and self._get_postive_prob() < self.positive_sample_ratio):
# make sure to crop the positive region
# the minimum top left to crop positive region (h,w)
tl = np.array([h + 1, w + 1], dtype=np.int32)
for gt_polygon in gt_polygons:
temp_point = np.min(gt_polygon.reshape(2, -1), axis=1)
if temp_point[0] <= tl[0]:
tl[0] = temp_point[0]
if temp_point[1] <= tl[1]:
tl[1] = temp_point[1]
tl = tl - (t_h, t_w)
tl[tl < 0] = 0
# the maximum bottum right to crop positive region
br = np.array([0, 0], dtype=np.int32)
for gt_polygon in gt_polygons:
temp_point = np.max(gt_polygon.reshape(2, -1), axis=1)
if temp_point[0] > br[0]:
br[0] = temp_point[0]
if temp_point[1] > br[1]:
br[1] = temp_point[1]
br = br - (t_h, t_w)
br[br < 0] = 0
# if br is too big so that crop the outside region of img
br[0] = min(br[0], h - t_h)
br[1] = min(br[1], w - t_w)
#
h = self._sample_num(tl[0], br[0]) if tl[0] < br[0] else 0
w = self._sample_num(tl[1], br[1]) if tl[1] < br[1] else 0
else:
# make sure not to crop outside of img
h = self._sample_num(0, h - t_h) if h - t_h > 0 else 0
w = self._sample_num(0, w - t_w) if w - t_w > 0 else 0
return (h, w)
def _crop_img(self, img: np.ndarray, offset: Tuple[int, int],
target_size: Tuple[int, int]) -> np.ndarray:
"""Crop the image given an offset and a target size.
Args:
img (ndarray): Image.
offset (Tuple[int. int]): Coordinates of the starting point.
target_size: Target image size.
"""
h, w = img.shape[:2]
target_size = target_size[::-1]
br = np.min(
np.stack((np.array(offset) + np.array(target_size), np.array(
(h, w)))),
axis=0)
return img[offset[0]:br[0], offset[1]:br[1]], np.array(
[offset[1], offset[0], br[1], br[0]])
def _crop_polygons(self, polygons: Sequence[np.ndarray],
crop_bbox: np.ndarray) -> Sequence[np.ndarray]:
"""Crop polygons to be within a crop region. If polygon crosses the
crop_bbox, we will keep the part left in crop_bbox by cropping its
boardline.
Args:
polygons (list(ndarray)): List of polygons [(N1, ), (N2, ), ...].
crop_bbox (ndarray): Cropping region. [x1, y1, x2, y1].
Returns
tuple(List(ArrayLike), list[int]):
- (List(ArrayLike)): The rest of the polygons located in the
crop region.
- (list[int]): Index list of the reserved polygons.
"""
polygons_cropped = []
kept_idx = []
for idx, polygon in enumerate(polygons):
if polygon.size < 6:
continue
poly = crop_polygon(polygon, crop_bbox)
if poly is not None:
poly = poly.reshape(-1, 2) - (crop_bbox[0], crop_bbox[1])
polygons_cropped.append(poly.reshape(-1))
kept_idx.append(idx)
return (polygons_cropped, kept_idx)
def transform(self, results: Dict) -> Dict:
"""Applying random crop on results.
Args:
results (dict): Result dict contains the data to transform.
Returns:
dict: The transformed data
"""
if self.target_size == results['img'].shape[:2][::-1]:
return results
gt_polygons = results['gt_polygons']
crop_offset = self._sample_offset(gt_polygons,
results['img'].shape[:2])
img, crop_bbox = self._crop_img(results['img'], crop_offset,
self.target_size)
results['img'] = img
results['img_shape'] = img.shape[:2]
gt_polygons, polygon_kept_idx = self._crop_polygons(
gt_polygons, crop_bbox)
bboxes = [poly2bbox(poly) for poly in gt_polygons]
results['gt_bboxes'] = np.array(
bboxes, dtype=np.float32).reshape(-1, 4)
results['gt_polygons'] = gt_polygons
results['gt_bboxes_labels'] = results['gt_bboxes_labels'][
polygon_kept_idx]
results['gt_ignored'] = results['gt_ignored'][polygon_kept_idx]
return results
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(target_size = {self.target_size}, '
repr_str += f'positive_sample_ratio = {self.positive_sample_ratio})'
return repr_str