File size: 20,599 Bytes
3d1f2c9 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 |
import copy
import torch
import numpy as np
from scipy.stats import linregress
from typing import List, Optional, Tuple
def generate_gaussian_matrix_vectorized(h, w, px, py, sigma=2):
# Create a grid of indices
x, y = np.meshgrid(np.arange(h), np.arange(w))
# Calculate Gaussian values for the entire grid
matrix = np.exp(-((x - px)**2 + (y - py)**2) / (2 * sigma**2))
return matrix
def resize_keypoints(keypoints, original_size, new_size):
ratio_h = new_size[0] / original_size[0]
ratio_w = new_size[1] / original_size[1]
resized_keypoints = {}
for kp, values in keypoints.items():
x_resized = int(values['x'] * ratio_w)
y_resized = int(values['y'] * ratio_h)
resized_keypoints[kp] = {'x': x_resized, 'y': y_resized, 'in_frame': values['in_frame']}
if 'proj_err' in values.keys():
resized_keypoints[kp]['proj_err'] = values['proj_err']
return resized_keypoints
def generate_gaussian_array_vectorized(num_matrices, keypoints, original_size, down_ratio=2, sigma=2, proj_err_th=5.):
new_size = tuple(ti/down_ratio for ti in original_size)
resized_keypoints = resize_keypoints(keypoints, original_size, new_size)
# Create an array of center points based on resized keypoints
center_points = []
for kp in range(1, num_matrices):
if kp in resized_keypoints.keys():
if (resized_keypoints[kp]['in_frame']):
if 'proj_err' in resized_keypoints[kp].keys():
if resized_keypoints[kp]['proj_err'] <= proj_err_th:
center_points.append([resized_keypoints[kp]['x'], resized_keypoints[kp]['y']])
else:
center_points.append([np.inf, np.inf])
else:
center_points.append([resized_keypoints[kp]['x'], resized_keypoints[kp]['y']])
else:
center_points.append([np.inf, np.inf])
else:
center_points.append([np.inf, np.inf])
center_points = np.array(center_points)
# Generate Gaussian matrices for all center points
matrices = [generate_gaussian_matrix_vectorized(new_size[0], new_size[1], px, py, sigma) for px, py in center_points]
matrices = np.array(matrices)
matrices = np.concatenate((matrices, 1-matrices.sum(axis=0, keepdims=True)), axis=0)
matrices = np.clip(matrices, 0, 1)
return matrices
def resize_keypoints_l(keypoints, original_size, new_size):
ratio_h = new_size[0] / original_size[0]
ratio_w = new_size[1] / original_size[1]
resized_keypoints = {}
for kp, values in keypoints.items():
x1_resized = int(values['x_1'] * ratio_w)
y1_resized = int(values['y_1'] * ratio_h)
x2_resized = int(values['x_2'] * ratio_w)
y2_resized = int(values['y_2'] * ratio_h)
resized_keypoints[kp] = {'x_1': x1_resized, 'y_1': y1_resized, 'x_2': x2_resized, 'y_2': y2_resized}
return resized_keypoints
def generate_gaussian_array_vectorized_l(num_matrices, keypoints, original_size, down_ratio=2, sigma=2, sigma_mult=1):
def sigma_f(px, py, size, sigma):
#multiply sigma if point in image border
if (px < 5 or px > size[0] - 5) | (py < 5 or py > size[1] - 5):
return sigma_mult*sigma
else:
return sigma
new_size = tuple(int(ti / down_ratio) for ti in original_size)
resized_keypoints = resize_keypoints_l(keypoints, original_size, new_size)
# Create an array of center points based on resized keypoints for both points
center_points = []
for kp in range(1, num_matrices+1):
if kp in resized_keypoints.keys():
center_points.append([resized_keypoints[kp]['x_1'], resized_keypoints[kp]['y_1']])
center_points.append([resized_keypoints[kp]['x_2'], resized_keypoints[kp]['y_2']])
else:
center_points.append([np.inf, np.inf])
center_points.append([np.inf, np.inf])
center_points = np.array(center_points)
# Generate Gaussian matrices for both points and sum them
matrices1 = [generate_gaussian_matrix_vectorized(new_size[0], new_size[1], px, py, sigma_f(px, py, new_size, sigma)) for px, py in
center_points[::2]]
matrices2 = [generate_gaussian_matrix_vectorized(new_size[0], new_size[1], px, py, sigma_f(px, py, new_size, sigma)) for px, py in
center_points[1::2]]
matrices = np.array(matrices1) + np.array(matrices2)
matrices_border = np.zeros((1, new_size[1], new_size[0]))
for kp in range(1, num_matrices+1):
if kp in resized_keypoints.keys():
x1, y1 = resized_keypoints[kp]['x_1'], resized_keypoints[kp]['y_1']
x2, y2 = resized_keypoints[kp]['x_2'], resized_keypoints[kp]['y_2']
pixel_dist = np.linalg.norm(np.array([x2, y2]) - np.array([x1, y1]))
num_gaussians = int(pixel_dist / (sigma))
if num_gaussians != 1:
for i in range(num_gaussians):
alpha = i / (num_gaussians - 1)
x = int(x1 + alpha * (x2 - x1))
y = int(y1 + alpha * (y2 - y1))
matrices_border[0, :, :] += generate_gaussian_matrix_vectorized(new_size[0], new_size[1], x, y, sigma)
else:
x, y = abs(x2 - x1) / 2, abs(y2 - y1) / 2
matrices_border[0, :, :] += generate_gaussian_matrix_vectorized(new_size[0], new_size[1], x, y, sigma)
matrices_border = np.clip(matrices_border, 0, 1)
matrices_combined = np.concatenate((matrices, matrices_border), axis=0)
return matrices_combined
def get_keypoints_from_heatmap_batch_maxpool(
heatmap: torch.Tensor,
scale: int = 2,
max_keypoints: int = 1,
min_keypoint_pixel_distance: int = 15,
return_scores: bool = True,
) -> List[List[List[Tuple[int, int]]]]:
"""Fast extraction of keypoints from a batch of heatmaps using maxpooling.
Inspired by mmdetection and CenterNet:
https://mmdetection.readthedocs.io/en/v2.13.0/_modules/mmdet/models/utils/gaussian_target.html
Args:
heatmap (torch.Tensor): NxCxHxW heatmap batch
max_keypoints (int, optional): max number of keypoints to extract, lowering will result in faster execution times. Defaults to 20.
min_keypoint_pixel_distance (int, optional): _description_. Defaults to 1.
Following thresholds can be used at inference time to select where you want to be on the AP curve. They should ofc. not be used for training
abs_max_threshold (Optional[float], optional): _description_. Defaults to None.
rel_max_threshold (Optional[float], optional): _description_. Defaults to None.
Returns:
The extracted keypoints for each batch, channel and heatmap; and their scores
"""
batch_size, n_channels, _, width = heatmap.shape
# obtain max_keypoints local maxima for each channel (w/ maxpool)
kernel = min_keypoint_pixel_distance * 2 + 1
pad = min_keypoint_pixel_distance
# exclude border keypoints by padding with highest possible value
# bc the borders are more susceptible to noise and could result in false positives
padded_heatmap = torch.nn.functional.pad(heatmap, (pad, pad, pad, pad), mode="constant", value=1.0)
max_pooled_heatmap = torch.nn.functional.max_pool2d(padded_heatmap, kernel, stride=1, padding=0)
# if the value equals the original value, it is the local maximum
local_maxima = max_pooled_heatmap == heatmap
# all values to zero that are not local maxima
heatmap = heatmap * local_maxima
# extract top-k from heatmap (may include non-local maxima if there are less peaks than max_keypoints)
scores, indices = torch.topk(heatmap.view(batch_size, n_channels, -1), max_keypoints, sorted=True)
indices = torch.stack([torch.div(indices, width, rounding_mode="floor"), indices % width], dim=-1)
# at this point either score > 0.0, in which case the index is a local maximum
# or score is 0.0, in which case topk returned non-maxima, which will be filtered out later.
# remove top-k that are not local maxima and threshold (if required)
# thresholding shouldn't be done during training
# moving them to CPU now to avoid multiple GPU-mem accesses!
indices = indices.detach().cpu().numpy()
scores = scores.detach().cpu().numpy()
filtered_indices = [[[] for _ in range(n_channels)] for _ in range(batch_size)]
filtered_scores = [[[] for _ in range(n_channels)] for _ in range(batch_size)]
# have to do this manually as the number of maxima for each channel can be different
for batch_idx in range(batch_size):
for channel_idx in range(n_channels):
candidates = indices[batch_idx, channel_idx]
locs = []
for candidate_idx in range(candidates.shape[0]):
# convert to (u,v)
loc = candidates[candidate_idx][::-1] * scale
loc = loc.tolist()
if return_scores:
loc.append(scores[batch_idx, channel_idx, candidate_idx])
locs.append(loc)
filtered_indices[batch_idx][channel_idx] = locs
return torch.tensor(filtered_indices)
def get_keypoints_from_heatmap_batch_maxpool_l(
heatmap: torch.Tensor,
scale: int = 2,
max_keypoints: int = 2,
min_keypoint_pixel_distance: int = 10,
return_scores: bool = True,
) -> List[List[List[Tuple[int, int]]]]:
"""Fast extraction of keypoints from a batch of heatmaps using maxpooling.
Inspired by mmdetection and CenterNet:
https://mmdetection.readthedocs.io/en/v2.13.0/_modules/mmdet/models/utils/gaussian_target.html
Args:
heatmap (torch.Tensor): NxCxHxW heatmap batch
max_keypoints (int, optional): max number of keypoints to extract, lowering will result in faster execution times. Defaults to 20.
min_keypoint_pixel_distance (int, optional): _description_. Defaults to 1.
Following thresholds can be used at inference time to select where you want to be on the AP curve. They should ofc. not be used for training
abs_max_threshold (Optional[float], optional): _description_. Defaults to None.
rel_max_threshold (Optional[float], optional): _description_. Defaults to None.
Returns:
The extracted keypoints for each batch, channel and heatmap; and their scores
"""
batch_size, n_channels, _, width = heatmap.shape
kernel = min_keypoint_pixel_distance * 2 + 1
pad = int((kernel-1)/2)
max_pooled_heatmap = torch.nn.functional.max_pool2d(heatmap, kernel, stride=1, padding=pad)
# if the value equals the original value, it is the local maximum
local_maxima = max_pooled_heatmap == heatmap
# all values to zero that are not local maxima
heatmap = heatmap * local_maxima
# extract top-k from heatmap (may include non-local maxima if there are less peaks than max_keypoints)
scores, indices = torch.topk(heatmap.view(batch_size, n_channels, -1), max_keypoints, sorted=True)
indices = torch.stack([torch.div(indices, width, rounding_mode="floor"), indices % width], dim=-1)
# at this point either score > 0.0, in which case the index is a local maximum
# or score is 0.0, in which case topk returned non-maxima, which will be filtered out later.
# remove top-k that are not local maxima and threshold (if required)
# thresholding shouldn't be done during training
# moving them to CPU now to avoid multiple GPU-mem accesses!
indices = indices.detach().cpu().numpy()
scores = scores.detach().cpu().numpy()
filtered_indices = [[[] for _ in range(n_channels)] for _ in range(batch_size)]
filtered_scores = [[[] for _ in range(n_channels)] for _ in range(batch_size)]
# have to do this manually as the number of maxima for each channel can be different
for batch_idx in range(batch_size):
for channel_idx in range(n_channels):
candidates = indices[batch_idx, channel_idx]
locs = []
for candidate_idx in range(candidates.shape[0]):
# convert to (u,v)
loc = candidates[candidate_idx][::-1] * scale
loc = loc.tolist()
if return_scores:
loc.append(scores[batch_idx, channel_idx, candidate_idx])
locs.append(loc)
filtered_indices[batch_idx][channel_idx] = locs
return torch.tensor(filtered_indices)
def coords_to_dict(coords, threshold=0.05, ground_plane_only=False):
kp_list = []
for batch in range(coords.size()[0]):
keypoints = {}
for count, c in enumerate(range(coords.size(1))):
if coords.size(2) == 1:
if ground_plane_only and c+1 in [12,15,16,19]:
continue
if coords[batch, c, 0, -1] > threshold:
keypoints[count+1] = {'x': coords[batch, c, 0, 0].item(),
'y': coords[batch, c, 0, 1].item(),
'p': coords[batch, c, 0, 2].item()}
else:
if ground_plane_only and c+1 in [7,8,9,10,11,12]:
continue
if coords[batch, c, 0, -1] > threshold and coords[batch, c, 1, -1] > threshold:
keypoints[count+1] = {'x_1': coords[batch, c, 0, 0].item(),
'y_1': coords[batch, c, 0, 1].item(),
'p_1': coords[batch, c, 0, 2].item(),
'x_2': coords[batch, c, 1, 0].item(),
'y_2': coords[batch, c, 1, 1].item(),
'p_2': coords[batch, c, 1, 2].item()}
kp_list.append(keypoints)
return kp_list
def complete_keypoints(kp_dict, lines_dict, w, h, normalize=False):
def line_intersection(x1, y1, x2, y2):
#1e-7 sum in case there are two identical coordinate values
x1[-1] += 1e-7
x2[-1] += 1e-7
slope1, intercept1, r1, p1, se1 = linregress(x1, y1)
slope2, intercept2, r2, p2, se2 = linregress(x2, y2)
x_intersection = (intercept2 - intercept1) / (slope1 - slope2 + 1e-7)
y_intersection = slope1 * x_intersection + intercept1
return x_intersection, y_intersection
lines_list = ["Big rect. left bottom", "Big rect. left main", "Big rect. left top", "Big rect. right bottom",
"Big rect. right main", "Big rect. right top", "Goal left crossbar", "Goal left post left ",
"Goal left post right", "Goal right crossbar", "Goal right post left", "Goal right post right",
"Middle line", "Side line bottom", "Side line left", "Side line right", "Side line top",
"Small rect. left bottom", "Small rect. left main", "Small rect. left top", "Small rect. right bottom",
"Small rect. right main", "Small rect. right top"]
keypoints_line_list = [['Side line top', 'Side line left'], ['Side line top', 'Middle line'],
['Side line right', 'Side line top'], ['Side line left', 'Big rect. left top'],
['Big rect. left top', 'Big rect. left main'], ['Big rect. right top', 'Big rect. right main'],
['Side line right', 'Big rect. right top'], ['Side line left', 'Small rect. left top'],
['Small rect. left top', 'Small rect. left main'], ['Small rect. right top', 'Small rect. right main'],
['Side line right', 'Small rect. right top'], ['Goal left crossbar', 'Goal left post right'],
['Side line left', 'Goal left post right'], ['Side line right', 'Goal right post left'],
['Goal right crossbar', 'Goal right post left'], ['Goal left crossbar', 'Goal left post left '],
['Side line left', 'Goal left post left '], ['Side line right', 'Goal right post right'],
['Goal right crossbar', 'Goal right post right'], ['Side line left', 'Small rect. left bottom'],
['Small rect. left bottom', 'Small rect. left main'], ['Small rect. right bottom', 'Small rect. right main'],
['Side line right', 'Small rect. right bottom'], ['Side line left', 'Big rect. left bottom'],
['Big rect. left bottom', 'Big rect. left main'], ['Big rect. right main', 'Big rect. right bottom'],
['Side line right', 'Big rect. right bottom'], ['Side line left', 'Side line bottom'],
['Side line bottom', 'Middle line'], ['Side line bottom', 'Side line right']]
keypoint_aux_pair_list = [['Small rect. left main', 'Side line top'], ['Big rect. left main', 'Side line top'],
['Big rect. right main', 'Side line top'], ['Small rect. right main', 'Side line top'],
['Small rect. left main', 'Big rect. left top'], ['Big rect. right top', 'Small rect. right main'],
['Small rect. left top', 'Big rect. left main'], ['Small rect. right top', 'Big rect. right main'],
['Small rect. left bottom', 'Big rect. left main'], ['Small rect. right bottom', 'Big rect. right main'],
['Small rect. left main', 'Big rect. left bottom'], ['Small rect. right main', 'Big rect. right bottom'],
['Small rect. left main', 'Side line bottom'], ['Big rect. left main', 'Side line bottom'],
['Big rect. right main', 'Side line bottom'], ['Small rect. right main', 'Side line bottom']]
w_extra = 0. * w
h_extra = 0. * h
complete_dict = copy.deepcopy(kp_dict)
for key in range(1, 31):
if key not in kp_dict.keys():
line_keys = keypoints_line_list[key-1]
line_key1, line_key2 = lines_list.index(line_keys[0]) + 1, lines_list.index(line_keys[1]) + 1
if all(line_key in lines_dict.keys() for line_key in [line_key1, line_key2]):
x1 = [lines_dict[line_key1]['x_1'], lines_dict[line_key1]['x_2']]
y1 = [lines_dict[line_key1]['y_1'], lines_dict[line_key1]['y_2']]
x2 = [lines_dict[line_key2]['x_1'], lines_dict[line_key2]['x_2']]
y2 = [lines_dict[line_key2]['y_1'], lines_dict[line_key2]['y_2']]
new_kp = line_intersection(x1, y1, x2, y2)
if -w_extra < new_kp[0] < w_extra + w and -h_extra < new_kp[1] < h_extra + h:
complete_dict[key] = {'x': round(new_kp[0], 0), 'y': round(new_kp[1], 0), 'p': 1.}
for key in range(1, len(keypoint_aux_pair_list)):
line_keys = keypoint_aux_pair_list[key-1]
line_key1, line_key2 = lines_list.index(line_keys[0]) + 1, lines_list.index(line_keys[1]) + 1
if all(line_key in lines_dict.keys() for line_key in [line_key1, line_key2]):
x1 = [lines_dict[line_key1]['x_1'], lines_dict[line_key1]['x_2']]
y1 = [lines_dict[line_key1]['y_1'], lines_dict[line_key1]['y_2']]
x2 = [lines_dict[line_key2]['x_1'], lines_dict[line_key2]['x_2']]
y2 = [lines_dict[line_key2]['y_1'], lines_dict[line_key2]['y_2']]
new_kp = line_intersection(x1, y1, x2, y2)
if -w_extra < new_kp[0] < w_extra + w and -h_extra < new_kp[1] < h_extra + h:
complete_dict[key+57] = {'x': round(new_kp[0], 0), 'y': round(new_kp[1], 0), 'p': 1.}
if normalize:
for kp in complete_dict.keys():
complete_dict[kp]['x'] /= w
complete_dict[kp]['y'] /= h
for line in lines_dict.keys():
lines_dict[line]['x_1'] /= w
lines_dict[line]['y_1'] /= h
lines_dict[line]['x_2'] /= w
lines_dict[line]['y_2'] /= h
complete_dict = dict(sorted(complete_dict.items()))
return complete_dict, lines_dict
|