Spaces:
Running
Running
File size: 3,293 Bytes
d8b3f49 |
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 |
# """
# Parts of the code are taken or adapted from
# https://github.com/mkocabas/EpipolarPose/blob/master/lib/utils/img_utils.py
# """
from typing import Tuple
import cv2
import numpy as np
def generate_image_patch_cv2_resize(
img: np.array,
c_x: float,
c_y: float,
bb_width: float,
bb_height: float,
patch_width: float,
patch_height: float,
do_flip: bool,
scale: float,
rot: float,
) -> Tuple[np.array, np.array]:
"""
Crop and resize the input image with zero padding when needed
"""
assert bb_width == bb_height
img_height, img_width, img_channels = img.shape
if do_flip:
img = img[:, ::-1, :]
c_x = img_width - c_x - 1
# Calculate scaled dimensions
scaled_size = max(bb_width, bb_height) * scale
# Calculate desired crop coordinates without clamping
half_size = scaled_size / 2
x1_desired = c_x - half_size
y1_desired = c_y - half_size
x2_desired = c_x + half_size
y2_desired = c_y + half_size
# Calculate padding amounts
pad_left = max(0, -x1_desired)
pad_right = max(0, x2_desired - img_width)
pad_top = max(0, -y1_desired)
pad_bottom = max(0, y2_desired - img_height)
# Create padded image
padded_height = img_height + int(pad_top + pad_bottom)
padded_width = img_width + int(pad_left + pad_right)
if img_channels == 4:
padded_img = np.zeros((padded_height, padded_width, 4), dtype=img.dtype)
else:
padded_img = np.zeros((padded_height, padded_width, img_channels), dtype=img.dtype)
# Copy original image into padded image
padded_img[
int(pad_top) : int(pad_top + img_height),
int(pad_left) : int(pad_left + img_width),
] = img
# Calculate crop coordinates in padded image
x1 = max(0, int(x1_desired + pad_left))
y1 = max(0, int(y1_desired + pad_top))
x2 = x1 + int(scaled_size)
y2 = y1 + int(scaled_size)
# Crop the square region
cropped = padded_img[y1:y2, x1:x2]
# Apply rotation if needed
if rot != 0:
center = (scaled_size / 2, scaled_size / 2)
rot_mat = cv2.getRotationMatrix2D(center, rot, 1.0)
cropped = cv2.warpAffine(cropped, rot_mat, (int(scaled_size), int(scaled_size)))
# Resize to target size
if img_channels == 4:
rgb = cv2.resize(
cropped[:, :, :3],
(int(patch_width), int(patch_height)),
interpolation=cv2.INTER_LINEAR,
)
alpha = cv2.resize(
cropped[:, :, 3],
(int(patch_width), int(patch_height)),
interpolation=cv2.INTER_LINEAR,
)
img_patch = np.dstack((rgb, alpha))
else:
img_patch = cv2.resize(
cropped,
(int(patch_width), int(patch_height)),
interpolation=cv2.INTER_LINEAR,
)
return img_patch
def convert_cvimg_to_tensor(cvimg: np.array):
"""
Convert image from HWC to CHW format.
Args:
cvimg (np.array): Image of shape (H, W, 3) as loaded by OpenCV.
Returns:
np.array: Output image of shape (3, H, W).
"""
# from h,w,c(OpenCV) to c,h,w
img = cvimg.copy()
img = np.transpose(img, (2, 0, 1))
# from int to float
img = img.astype(np.float32)
return img
|