File size: 13,064 Bytes
25986db | 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 | import torch
import math
import cv2 as cv
import torch.nn.functional as F
import numpy as np
'''modified from the original test implementation
Replace cv.BORDER_REPLICATE with cv.BORDER_CONSTANT
Add a variable called att_mask for computing attention and positional encoding later'''
def sample_target(im, target_bb, search_area_factor, output_sz=None, mask=None):
""" Extracts a square crop centered at target_bb box, of area search_area_factor^2 times target_bb area
args:
im - cv image
target_bb - target box [x, y, w, h]
search_area_factor - Ratio of crop size to target size
output_sz - (float) Size to which the extracted crop is resized (always square). If None, no resizing is done.
returns:
cv image - extracted crop
float - the factor by which the crop has been resized to make the crop size equal output_size
"""
if not isinstance(target_bb, list):
x, y, w, h = target_bb.tolist()
else:
x, y, w, h = target_bb
# Crop image
crop_sz = math.ceil(math.sqrt(w * h) * search_area_factor)
if crop_sz < 1:
raise Exception('Too small bounding box.')
x1 = round(x + 0.5 * w - crop_sz * 0.5)
x2 = x1 + crop_sz
y1 = round(y + 0.5 * h - crop_sz * 0.5)
y2 = y1 + crop_sz
x1_pad = max(0, -x1)
x2_pad = max(x2 - im.shape[1] + 1, 0)
y1_pad = max(0, -y1)
y2_pad = max(y2 - im.shape[0] + 1, 0)
# Crop target
im_crop = im[y1 + y1_pad:y2 - y2_pad, x1 + x1_pad:x2 - x2_pad, :]
if mask is not None:
mask_crop = mask[y1 + y1_pad:y2 - y2_pad, x1 + x1_pad:x2 - x2_pad]
# Pad
im_crop_padded = cv.copyMakeBorder(im_crop, y1_pad, y2_pad, x1_pad, x2_pad, cv.BORDER_CONSTANT)
# deal with attention mask
H, W, _ = im_crop_padded.shape
att_mask = np.ones((H,W))
end_x, end_y = -x2_pad, -y2_pad
if y2_pad == 0:
end_y = None
if x2_pad == 0:
end_x = None
att_mask[y1_pad:end_y, x1_pad:end_x] = 0
if mask is not None:
mask_crop_padded = F.pad(mask_crop, pad=(x1_pad, x2_pad, y1_pad, y2_pad), mode='constant', value=0)
if output_sz is not None:
resize_factor = output_sz / crop_sz
im_crop_padded = cv.resize(im_crop_padded, (output_sz, output_sz))
att_mask = cv.resize(att_mask, (output_sz, output_sz)).astype(np.bool_)
if mask is None:
return im_crop_padded, resize_factor, att_mask
mask_crop_padded = \
F.interpolate(mask_crop_padded[None, None], (output_sz, output_sz), mode='bilinear', align_corners=False)[0, 0]
return im_crop_padded, resize_factor, att_mask, mask_crop_padded
else:
if mask is None:
return im_crop_padded, att_mask.astype(np.bool_), 1.0
return im_crop_padded, 1.0, att_mask.astype(np.bool_), mask_crop_padded
def resize_sample_target(im, target_bb, output_sz=None, mask=None):
""" Resize the image
args:
im - cv image
target_bb - target box [x, y, w, h]
output_sz - (float) Size to which the extracted crop is resized (always square). If None, no resizing is done.
returns:
cv image - extracted crop
float - the factor by which the crop has been resized to make the crop size equal output_size
"""
# Resize image
# deal with attention mask
H, W, _ = im.shape
att_mask = np.zeros((H,W))
if output_sz is not None:
resize_factor = (output_sz / W, output_sz / H) # (w,h) rather than (h,w)
im_resized = cv.resize(im, (output_sz, output_sz))
att_mask = cv.resize(att_mask, (output_sz, output_sz)).astype(np.bool_)
if mask is None:
return im_resized, resize_factor, att_mask
mask_resized = \
F.interpolate(mask[None, None], (output_sz, output_sz), mode='bilinear', align_corners=False)[0, 0]
return im_resized, resize_factor, att_mask, mask_resized
else:
if mask is None:
return im, att_mask.astype(np.bool_), 1.0
return im, 1.0, att_mask.astype(np.bool_), mask
def transform_image_to_crop(box_in: torch.Tensor, box_extract: torch.Tensor, resize_factor: float,
crop_sz: torch.Tensor, normalize=False) -> torch.Tensor:
""" Transform the box co-ordinates from the original image co-ordinates to the co-ordinates of the cropped image
args:
box_in - the box for which the co-ordinates are to be transformed
box_extract - the box about which the image crop has been extracted.
resize_factor - the ratio between the original image scale and the scale of the image crop
crop_sz - size of the cropped image
returns:
torch.Tensor - transformed co-ordinates of box_in
"""
box_extract_center = box_extract[0:2] + 0.5 * box_extract[2:4]
box_in_center = box_in[0:2] + 0.5 * box_in[2:4]
box_out_center = (crop_sz - 1) / 2 + (box_in_center - box_extract_center) * resize_factor
box_out_wh = box_in[2:4] * resize_factor
box_out = torch.cat((box_out_center - 0.5 * box_out_wh, box_out_wh))
if normalize:
#crop_sz[0] - 1, modified by chenxin from crop_sz[0],2022.7.15
return box_out / (crop_sz[0]-1)
else:
return box_out
def transform_image_to_resize(box_in: torch.Tensor, resize_factor: float,
crop_sz: torch.Tensor, normalize=False) -> torch.Tensor:
""" Transform the box co-ordinates from the original image co-ordinates to the co-ordinates of the cropped image
args:
box_in - the box for which the co-ordinates are to be transformed
box_extract - the box about which the image crop has been extracted.
resize_factor - the ratio between the original image scale and the scale of the image crop
crop_sz - size of the cropped image
returns:
torch.Tensor - transformed co-ordinates of box_in
"""
box_out_xy = box_in[:2] * torch.tensor(resize_factor)
box_out_wh = box_in[2:4] * torch.tensor(resize_factor)
box_out = torch.cat((box_out_xy, box_out_wh))
if normalize:
return box_out / (crop_sz[0]-1)
else:
return box_out
def jittered_center_crop(frames, box_extract, box_gt, search_area_factor, output_sz, masks=None):
""" For each frame in frames, extracts a square crop centered at box_extract, of area search_area_factor^2
times box_extract area. The extracted crops are then resized to output_sz. Further, the co-ordinates of the box
box_gt are transformed to the image crop co-ordinates
args:
frames - list of frames
box_extract - list of boxes of same length as frames. The crops are extracted using anno_extract
box_gt - list of boxes of same length as frames. The co-ordinates of these boxes are transformed from
image co-ordinates to the crop co-ordinates
search_area_factor - The area of the extracted crop is search_area_factor^2 times box_extract area
output_sz - The size to which the extracted crops are resized
returns:
list - list of image crops
list - box_gt location in the crop co-ordinates
"""
if masks is None:
crops_resize_factors = [sample_target(f, a, search_area_factor, output_sz)
for f, a in zip(frames, box_extract)]
frames_crop, resize_factors, att_mask = zip(*crops_resize_factors)
masks_crop = None
else:
crops_resize_factors = [sample_target(f, a, search_area_factor, output_sz, m)
for f, a, m in zip(frames, box_extract, masks)]
frames_crop, resize_factors, att_mask, masks_crop = zip(*crops_resize_factors)
# frames_crop: tuple of ndarray (128,128,3), att_mask: tuple of ndarray (128,128)
crop_sz = torch.Tensor([output_sz, output_sz])
# find the bb location in the crop
'''Note that here we use normalized coord'''
box_crop = [transform_image_to_crop(a_gt, a_ex, rf, crop_sz, normalize=True)
for a_gt, a_ex, rf in zip(box_gt, box_extract, resize_factors)] # (x1,y1,w,h) list of tensors
return frames_crop, box_crop, att_mask, masks_crop
def pstb_jittered_center_crop(frames, box_extract, box_gt, box_frame, search_area_factor, output_sz, masks=None):
""" For each frame in frames, extracts a square crop centered at box_extract, of area search_area_factor^2
times box_extract area. The extracted crops are then resized to output_sz. Further, the co-ordinates of the box
box_gt are transformed to the image crop co-ordinates
args:
frames - list of frames
box_extract - list of boxes of same length as frames. The crops are extracted using anno_extract
box_gt - list of boxes of same length as frames. The co-ordinates of these boxes are transformed from
image co-ordinates to the crop co-ordinates
search_area_factor - The area of the extracted crop is search_area_factor^2 times box_extract area
output_sz - The size to which the extracted crops are resized
returns:
list - list of image crops
list - box_gt location in the crop co-ordinates
"""
if masks is None:
crops_resize_factors = [sample_target(f, a, search_area_factor, output_sz)
for f, a in zip(frames, box_extract)]
frames_crop, resize_factors, att_mask = zip(*crops_resize_factors)
masks_crop = None
else:
crops_resize_factors = [sample_target(f, a, search_area_factor, output_sz, m)
for f, a, m in zip(frames, box_extract, masks)]
frames_crop, resize_factors, att_mask, masks_crop = zip(*crops_resize_factors)
# frames_crop: tuple of ndarray (128,128,3), att_mask: tuple of ndarray (128,128)
crop_sz = torch.Tensor([output_sz, output_sz])
# find the bb location in the crop
'''Note that here we use normalized coord'''
box_crop = [transform_image_to_crop(a_gt, a_ex, rf, crop_sz, normalize=True)
for a_gt, a_ex, rf in zip(box_gt, box_extract, resize_factors)] # (x1,y1,w,h) list of tensors
box_frame_crop = [transform_image_to_crop(a_gt, box_extract[-1], resize_factors[-1], crop_sz, normalize=True)
for a_gt in box_frame]
return frames_crop, box_crop, box_frame_crop, att_mask, masks_crop
def resize(frames, box, output_sz, masks=None):
""" For each frame in frames, extracts a square crop centered at box_extract, of area search_area_factor^2
times box_extract area. The extracted crops are then resized to output_sz. Further, the co-ordinates of the box
box_gt are transformed to the image crop co-ordinates
args:
frames - list of frames
box_extract - list of boxes of same length as frames. The crops are extracted using anno_extract
box_gt - list of boxes of same length as frames. The co-ordinates of these boxes are transformed from
image co-ordinates to the crop co-ordinates
search_area_factor - The area of the extracted crop is search_area_factor^2 times box_extract area
output_sz - The size to which the extracted crops are resized
returns:
list - list of image crops
list - box_gt location in the crop co-ordinates
"""
if masks is None:
crops_resize_factors = [resize_sample_target(f, a, output_sz)
for f, a in zip(frames, box)]
frames_crop, resize_factors, att_mask = zip(*crops_resize_factors)
masks_crop = None
else:
crops_resize_factors = [resize_sample_target(f, a, output_sz, m)
for f, a, m in zip(frames, box, masks)]
frames_crop, resize_factors, att_mask, masks_crop = zip(*crops_resize_factors)
# frames_crop: tuple of ndarray (128,128,3), att_mask: tuple of ndarray (128,128)
crop_sz = torch.Tensor([output_sz, output_sz])
# find the bb location in the crop
'''Note that here we use normalized coord'''
box_crop = [transform_image_to_resize(bb, rf, crop_sz, normalize=True)
for bb, rf in zip(box, resize_factors)] # (x1,y1,w,h) list of tensors
return frames_crop, box_crop, att_mask, masks_crop
def transform_box_to_crop(box: torch.Tensor, crop_box: torch.Tensor, crop_sz: torch.Tensor, normalize=False) -> torch.Tensor:
""" Transform the box co-ordinates from the original image co-ordinates to the co-ordinates of the cropped image
args:
box - the box for which the co-ordinates are to be transformed
crop_box - bounding box defining the crop in the original image
crop_sz - size of the cropped image
returns:
torch.Tensor - transformed co-ordinates of box_in
"""
box_out = box.clone()
box_out[:2] -= crop_box[:2]
scale_factor = crop_sz / crop_box[2:]
box_out[:2] *= scale_factor
box_out[2:] *= scale_factor
if normalize:
return box_out / crop_sz[0]
else:
return box_out
|