Spaces:
Running
Running
File size: 18,101 Bytes
e3641b1 5079645 e3641b1 4daa026 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 4daa026 e3641b1 4daa026 e3641b1 4daa026 e3641b1 4daa026 e3641b1 4daa026 5079645 4daa026 5079645 4daa026 5079645 4daa026 5079645 4daa026 e3641b1 4daa026 e3641b1 4daa026 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 e3641b1 5079645 4daa026 5079645 4daa026 5079645 4daa026 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 |
import cv2
import numpy as np
import matplotlib.pyplot as plt
from dtaidistance import dtw
from easy_ViTPose.inference import VitInference
import os
import requests
from pathlib import Path
from datetime import timedelta
from scipy.signal import savgol_filter
from scipy.stats import mstats
def predict_keypoints_vitpose(
video_path,
model_path,
model_name,
detector_path,
display_video=False
):
model = VitInference(
model=model_path,
yolo=detector_path,
model_name=model_name,
det_class=None,
dataset=None,
yolo_size=320,
is_video=False,
single_pose=False,
yolo_step=1
)
cap = cv2.VideoCapture(video_path)
detection_results = []
while True:
ret, frame = cap.read()
if not ret:
print(f"Keypoints were extracted from {video_path}")
break
frame = cv2.resize(frame, (1280, 720))
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
frame_keypoints = model.inference(frame)
if 0 in frame_keypoints:
detection_results.append(frame_keypoints[0])
if display_video:
frame = model.draw(False, False, 0.5)[..., ::-1]
if display_video:
cv2.imshow('preview', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
if display_video:
cap.release()
cv2.destroyAllWindows()
return np.array(detection_results)
def get_point_list_vitpose(detection_results):
return np.array(detection_results)[:, :, :-1]
def get_edge_groups(connections):
all_pairs = []
for i in range(len(connections)):
pairs = []
init_con = connections[i]
for k in range(len(connections)):
if k == i:
pass
candidat_con = connections[k]
point_1_init, point_2_init = init_con
point_1_candidat, point_2_candidat = candidat_con
if point_1_candidat == point_1_init or point_1_candidat == point_2_init or point_2_candidat == point_1_init or point_2_candidat == point_2_init:
pairs.append([init_con, candidat_con])
all_pairs.append(pairs)
all_point_for_edges = []
for set_of_pairs in all_pairs:
clean_pairs = []
for pair in set_of_pairs:
pair_a = pair[0]
pair_b = pair[1]
if len(list(set(pair_a + pair_b))) == 3:
center = int(list(set(pair_a) & set(pair_b))[0])
edges = list(set(pair_a) ^ set(pair_b))
points_for_edge = [edges[0], center, edges[1]]
clean_pairs.append(points_for_edge)
all_point_for_edges.extend(clean_pairs)
unique_set = set()
unique_list = []
for sublist in all_point_for_edges:
sublist_tuple = tuple(sublist)
if sublist_tuple not in unique_set:
unique_set.add(sublist_tuple)
unique_list.append(sublist)
unique_list.sort()
return unique_list
def calculate_angle(A, B, C):
A = np.round(np.array(A), decimals=3)
B = np.round(np.array(B), decimals=3)
C = np.round(np.array(C), decimals=3)
BA = A - B
BC = C - B
cosine_angle = np.dot(BA, BC) / ((np.linalg.norm(BA) * np.linalg.norm(BC)))
cosine_angle = np.clip(cosine_angle, -1, 1)
angle = np.arccos(cosine_angle)
if np.isnan(angle):
print(f"Invalid angle calculation.\n{A} \n{B} \n{C}")
minimum = np.min(np.array((np.linalg.norm(BA), np.linalg.norm(BC))))
return np.degrees(angle), minimum
def compute_all_angels(keypoints, edge_groups):
all_angles = []
for group in edge_groups:
A = keypoints[group[0]]
B = keypoints[group[1]]
C = keypoints[group[2]]
angle, minimum = calculate_angle(A, B, C)
all_angles.append([angle, minimum])
return np.array(all_angles)
def xy2phi(points_result, connections):
edge_groups = get_edge_groups(connections)
new_array = np.zeros((points_result.shape[0], len(edge_groups), 1))
for idx, frame in enumerate(points_result):
all_angels = compute_all_angels(keypoints=frame, edge_groups=edge_groups)[:, 0]
new_array[idx, :, :] = all_angels.reshape((len(edge_groups), 1))
return new_array
def get_series(point_list, edge_groups):
list_of_series = []
for edge_group in edge_groups:
keypoint_1, keypoint_2, keypoint_3 = edge_group
relevant_point_list = point_list[:, (keypoint_1, keypoint_2, keypoint_3), :]
series = []
for frame in relevant_point_list:
angle, _ = calculate_angle(frame[0, :], frame[1, :], frame[2, :])
series.append(angle)
list_of_series.append(series)
return np.array(list_of_series)
def plot_serieses(series_1, series_2):
plt.figure(dpi=150, figsize=(12, 5))
plt.plot(series_1, label='Video #1', lw=1)
plt.plot(series_2, label='Video #2', lw=1)
plt.axis("on")
plt.grid(True)
plt.xlabel("frames")
plt.ylabel("angles")
plt.legend()
def z_score_normalization(serieses, axis_for_znorm=1):
serieses_mean = np.mean(serieses, axis=axis_for_znorm, keepdims=True)
serieses_std = np.std(serieses, axis=axis_for_znorm, keepdims=True)
serieses_normalized = (serieses - serieses_mean) / serieses_std
return serieses_normalized
def get_dtw_mean_path(serieses_teacher, serieses_student, dtw_mean, dtw_filter):
list_of_paths = []
for idx in range(len(serieses_teacher)):
series_teacher = np.array(serieses_teacher[idx])
series_student = np.array(serieses_student[idx])
_ , paths = dtw.warping_paths(series_teacher, series_student, window=50)
path = dtw.best_path(paths)
list_of_paths.append(path)
all_dtw_tupples = []
for path in list_of_paths:
all_dtw_tupples.extend(path)
mean_path = []
for student_frame in range(len(serieses_student[0])):
frame_from_teacher = []
for frame_teacher in all_dtw_tupples:
if frame_teacher[1] == student_frame:
frame_from_teacher.append(frame_teacher[0])
mean_path.append((int(mstats.winsorize(np.array(frame_from_teacher), limits=[dtw_mean, dtw_mean]).mean()), student_frame))
path_array = np.array(mean_path)
smoothed_data = savgol_filter(path_array, window_length=dtw_filter, polyorder=0, axis=0)
path_array = np.array(smoothed_data).astype(int)
alignments = np.unique(path_array, axis=0) # TODO check if this correct
return alignments
def modify_student_frame(
detection_result_teacher,
detection_result_student,
detection_result_teacher_angles,
detection_result_student_angles,
video_teacher,
video_student,
alignment_frames,
edge_groups,
connections,
thresholds,
previously_trigered,
previously_trigered_2,
triger_state,
show_arrows,
text_dictionary,
):
arrows_bgr = (175, 75, 190)
arrows_sz = 3
skeleton_bgr = (0, 0, 255)
skeleton_sz = 3
frame_copy = video_student[alignment_frames[1]]
frame_teacher_copy = video_teacher[alignment_frames[0]]
frame_errors = np.abs(detection_result_teacher_angles[alignment_frames[0]] - detection_result_student_angles[alignment_frames[1]])
edge_groups_as_keys = [tuple(group) for group in edge_groups]
edge_groups2errors = dict(zip(edge_groups_as_keys, frame_errors))
edge_groups2thresholds = dict(zip(edge_groups_as_keys, thresholds))
edge_groups_relevant = [edge_group[1:] for edge_group in edge_groups]
text_info = []
trigered_connections = []
trigered_connections2 = []
for connection in connections:
edges_for_given_connection = [edge for edge in edge_groups2errors if connection[0] in edge or connection[1] in edge]
for edge in edges_for_given_connection:
check_threshold = edge_groups2errors[edge] > edge_groups2thresholds[edge]
check_certain = True
for keypoint in edge:
prob = detection_result_student[:, :,-1][alignment_frames[1]][keypoint]
if prob < 0.7:
check_certain = False
relevant_plane = [connection[0], connection[1]] in edge_groups_relevant or [connection[1], connection[0]] in edge_groups_relevant
if check_threshold and check_certain and relevant_plane:
point1, point2, point2_t = align_points(
detection_result_student,
detection_result_teacher,
alignment_frames,
edge
)
arrow = get_arrow_direction(point2, point2_t)
if triger_state == "one":
_ = cv2.line(frame_copy, point1, point2, skeleton_bgr, skeleton_sz)
if show_arrows:
_ = cv2.arrowedLine(frame_copy, point2, point2_t, arrows_bgr, arrows_sz)
if (connection[0], connection[1]) in text_dictionary:
text_info.append((text_dictionary[(connection[0], connection[1])], arrow))
if (connection[1], connection[0]) in text_dictionary:
text_info.append((text_dictionary[(connection[1], connection[0])], arrow))
if triger_state == "two":
trigered_connections.append((connection[0], connection[1]))
if (connection[0], connection[1]) in previously_trigered:
_ = cv2.line(frame_copy, point1, point2, skeleton_bgr, skeleton_sz)
if show_arrows:
_ = cv2.arrowedLine(frame_copy, point2, point2_t, arrows_bgr, arrows_sz)
if (connection[0], connection[1]) in text_dictionary:
text_info.append((text_dictionary[(connection[0], connection[1])], arrow))
if (connection[1], connection[0]) in text_dictionary:
text_info.append((text_dictionary[(connection[1], connection[0])], arrow))
if triger_state == "three":
trigered_connections.append((connection[0], connection[1]))
if (connection[0], connection[1]) in previously_trigered:
trigered_connections2.append((connection[0], connection[1]))
if (connection[0], connection[1]) in previously_trigered_2:
_ = cv2.line(frame_copy, point1, point2, skeleton_bgr, skeleton_sz)
if show_arrows:
_ = cv2.arrowedLine(frame_copy, point2, point2_t, arrows_bgr, arrows_sz)
if (connection[0], connection[1]) in text_dictionary:
text_info.append((text_dictionary[(connection[0], connection[1])], arrow))
if (connection[1], connection[0]) in text_dictionary:
text_info.append((text_dictionary[(connection[1], connection[0])], arrow))
return frame_copy, frame_teacher_copy, list(set(trigered_connections)), list(set(trigered_connections2)), text_info
def get_video_frames(video_path):
cap = cv2.VideoCapture(video_path)
video = []
while True:
ret, frame = cap.read()
if not ret:
print(f"Video {video_path} was loaded")
break
frame = cv2.resize(frame, (1280, 720))
video.append(frame)
return np.array(video)
def download_file(url, save_path):
response = requests.get(url, stream=True)
response.raise_for_status()
with open(save_path, 'wb') as file:
for chunk in response.iter_content(chunk_size=8192):
file.write(chunk)
def check_and_download_models():
# vit_model_s_url = "https://huggingface.co/JunkyByte/easy_ViTPose/resolve/main/torch/wholebody/vitpose-s-wholebody.pth?download=true"
vit_model_b_url = "https://huggingface.co/JunkyByte/easy_ViTPose/resolve/main/torch/wholebody/vitpose-b-wholebody.pth?download=true"
# vit_model_l_url = "https://huggingface.co/JunkyByte/easy_ViTPose/resolve/main/torch/wholebody/vitpose-l-wholebody.pth?download=true"
yolo_model_url = "https://huggingface.co/JunkyByte/easy_ViTPose/resolve/main/yolov8/yolov8s.pt?download=true"
# vit_model_s_path = "models/vitpose-s-wholebody.pth"
vit_model_b_path = "models/vitpose-b-wholebody.pth"
# vit_model_l_path = "models/vitpose-l-wholebody.pth"
yolo_model_path = "models/yolov8s.pt"
# Path(os.path.dirname(vit_model_s_path)).mkdir(parents=True, exist_ok=True)
Path(os.path.dirname(vit_model_b_path)).mkdir(parents=True, exist_ok=True)
# Path(os.path.dirname(vit_model_l_path)).mkdir(parents=True, exist_ok=True)
Path(os.path.dirname(yolo_model_path)).mkdir(parents=True, exist_ok=True)
# if not os.path.exists(vit_model_s_path):
# print("Downloading ViT-Pose-s model...")
# download_file(vit_model_s_url, vit_model_s_path)
# print("ViT-Pose-s model was downloaded.")
if not os.path.exists(vit_model_b_path):
print("Downloading ViT-Pose-b model...")
download_file(vit_model_b_url, vit_model_b_path)
print("ViT-Pose-b model was downloaded.")
# if not os.path.exists(vit_model_l_path):
# print("Downloading ViT-Pose-l model...")
# download_file(vit_model_l_url, vit_model_l_path)
# print("ViT-Pose-l model was downloaded.")
if not os.path.exists(yolo_model_path):
print("Downloading YOLO model...")
download_file(yolo_model_url, yolo_model_path)
print("YOLO model was downloaded.")
def generate_output_video(teacher_frames, student_frames, timestamp_str):
teacher_frames = np.array(teacher_frames)
student_frames = np.array(student_frames)
teacher_frames_resized = np.array([cv2.resize(frame, (1280, 720)) for frame in teacher_frames])
student_frames_resized = np.array([cv2.resize(frame, (1280, 720)) for frame in student_frames])
concat_video = np.concatenate((teacher_frames_resized, student_frames_resized), axis=2)
concat_video = np.array(concat_video)
root_dir = "videos"
if not os.path.exists(root_dir):
os.makedirs(root_dir)
video_path = f"{root_dir}/pose_{timestamp_str}.mp4"
out = cv2.VideoWriter(video_path, cv2.VideoWriter_fourcc(*'mp4v'), 30, (1280 * 2, 720))
for frame in concat_video:
out.write(frame)
out.release()
return video_path
def generate_log(all_text_summaries):
all_text_summaries_clean = list(set(all_text_summaries))
all_text_summaries_clean.sort(key=lambda x: x[1])
general_summary = []
for log in all_text_summaries_clean:
comment, frame, arrow = log
total_seconds = frame / 30
general_summary.append(f"{comment}. Direction: {arrow}. Video time: {str(timedelta(seconds=total_seconds))[3:-4]}")
general_summary = "\n".join(general_summary)
return general_summary
def write_log(
timestamp_str,
dtw_mean,
dtw_filter,
angles_sensitive,
angles_common,
angles_insensitive,
trigger_state,
general_summary
):
logs_dir = "logs"
if not os.path.exists(logs_dir):
os.makedirs(logs_dir)
log_path = f"{logs_dir}/log_{timestamp_str}.txt"
content = f"""
Settings:
Dynamic Time Warping:
- Winsorize mean: {dtw_mean}
- Savitzky-Golay Filter: {dtw_filter}
Thresholds:
- Sensitive: {angles_sensitive}
- Standart: {angles_common}
- Insensitive: {angles_insensitive}
Patience:
- trigger count: {trigger_state}
Error logs:
{general_summary}
"""
with open(log_path, "w") as file:
file.write(content)
print(f"log {log_path} was created.")
return log_path
def angle_between(v1, v2):
return np.arctan2(v2[1], v2[0]) - np.arctan2(v1[1], v1[0])
def align_points(detection_result_student, detection_result_teacher, alignment_frames, edge):
point0 = detection_result_student[alignment_frames[1], edge[0], :-1].astype(int)[::-1]
point1 = detection_result_student[alignment_frames[1], edge[1], :-1].astype(int)[::-1]
point2 = detection_result_student[alignment_frames[1], edge[2], :-1].astype(int)[::-1]
point0_t = detection_result_teacher[alignment_frames[0], edge[0], :-1].astype(int)[::-1]
point1_t = detection_result_teacher[alignment_frames[0], edge[1], :-1].astype(int)[::-1]
point2_t = detection_result_teacher[alignment_frames[0], edge[2], :-1].astype(int)[::-1]
translation = point0 - point0_t
point0_t += translation
point1_t += translation
point2_t += translation
BsA = point1 - point0
BtA = point1_t - point0
theta = angle_between(BtA, BsA)
R = np.array([
[np.cos(theta), -np.sin(theta)],
[np.sin(theta), np.cos(theta)]
])
point1_t = np.dot(R, (point1_t - point0).T).T + point0
point2_t = np.dot(R, (point2_t - point0).T).T + point0
point2_t = point2_t.astype(int)
return point1, point2, point2_t
def get_arrow_direction(A, B):
translation_vector = B - A
angle_deg = np.degrees(np.arctan2(translation_vector[0], translation_vector[1]))
match angle_deg:
case angle if -22.5 <= angle < 22.5:
arrow = "⬆"
case angle if 22.5 <= angle < 67.5:
arrow = "⬈"
case angle if 67.5 <= angle < 112.5:
arrow = "➡"
case angle if 112.5 <= angle < 157.5:
arrow = "⬊"
case angle if 157.5 <= angle or angle < -157.5:
arrow = "⬇"
case angle if -157.5 <= angle < -112.5:
arrow = "⬋"
case angle if -112.5 <= angle < -67.5:
arrow = "⬅"
case angle if -67.5 <= angle < -22.5:
arrow = "⬉"
case _:
arrow = ""
return arrow |