File size: 63,611 Bytes
6332c2a ed387b2 efa906f ed387b2 9bbcc8b 6332c2a ef2f1f1 ed387b2 9bbcc8b 4235cc6 9bbcc8b ef2f1f1 6332c2a ef2f1f1 6332c2a 9bbcc8b ef2f1f1 6332c2a ef2f1f1 6332c2a ef2f1f1 efa906f ef2f1f1 6332c2a ef2f1f1 6332c2a ef2f1f1 6332c2a ef2f1f1 efa906f 403bc03 ef2f1f1 403bc03 ef2f1f1 6332c2a ef2f1f1 ed387b2 ef2f1f1 403bc03 ef2f1f1 403bc03 ed387b2 ef2f1f1 ed387b2 ef2f1f1 403bc03 ed387b2 403bc03 ed387b2 403bc03 ef2f1f1 403bc03 ed387b2 403bc03 7f32c66 403bc03 ed387b2 ef2f1f1 ed387b2 7f32c66 ed387b2 ef2f1f1 7f32c66 403bc03 6332c2a 9bbcc8b 4235cc6 ed387b2 4235cc6 ed387b2 9bbcc8b 4235cc6 ed387b2 4235cc6 ed387b2 9bbcc8b ed387b2 4235cc6 ed387b2 9bbcc8b 4235cc6 9bbcc8b 4235cc6 ef2f1f1 4235cc6 ed387b2 4235cc6 403bc03 ef2f1f1 4235cc6 ef2f1f1 4235cc6 9bbcc8b ef2f1f1 4235cc6 ef2f1f1 9bbcc8b ef2f1f1 9bbcc8b 4235cc6 ed387b2 4235cc6 ef2f1f1 403bc03 ed387b2 403bc03 ed387b2 4235cc6 403bc03 ef2f1f1 4235cc6 ed387b2 ef2f1f1 ed387b2 4235cc6 ef2f1f1 4235cc6 ef2f1f1 ed387b2 4235cc6 403bc03 ed387b2 4235cc6 ef2f1f1 ed387b2 4235cc6 ef2f1f1 ed387b2 ef2f1f1 9bbcc8b ef2f1f1 4235cc6 ef2f1f1 4235cc6 ef2f1f1 ed387b2 ef2f1f1 9bbcc8b ef2f1f1 4235cc6 ef2f1f1 4235cc6 ef2f1f1 4235cc6 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 |
# import os
# import json
# import torch
# import numpy as np
# import gradio as gr
# import opts_egtea as opts
# from dataset import VideoDataSet, calc_iou
# from models import MYNET, SuppressNet
# from loss_func import cls_loss_func, regress_loss_func
# from eval import evaluation_detection
# from iou_utils import non_max_suppression, check_overlap_proposal
# from typing import List, Dict, Optional
# # Configuration
# VIS_CONFIG = {
# 'iou_threshold': 0.3,
# 'min_segment_duration': 1.0,
# }
# # Determine device
# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# print(f"Using device: {device}")
# def eval_frame(opt, model, dataset):
# """Evaluate model frame by frame"""
# test_loader = torch.utils.data.DataLoader(
# dataset,
# batch_size=opt['batch_size'],
# shuffle=False,
# num_workers=0,
# pin_memory=False
# )
# labels_cls = {video_name: [] for video_name in dataset.video_list}
# labels_reg = {video_name: [] for video_name in dataset.video_list}
# output_cls = {video_name: [] for video_name in dataset.video_list}
# output_reg = {video_name: [] for video_name in dataset.video_list}
# model.eval()
# with torch.no_grad():
# for n_iter, batch_data in enumerate(test_loader):
# try:
# if len(batch_data) == 4:
# input_data, cls_label, reg_label, _ = batch_data
# else:
# input_data, cls_label, reg_label = batch_data
# input_data = input_data.to(device)
# cls_label = cls_label.to(device) if cls_label is not None else None
# reg_label = reg_label.to(device) if reg_label is not None else None
# act_cls, act_reg, _ = model(input_data.float())
# act_cls = torch.softmax(act_cls, dim=-1)
# for b in range(input_data.size(0)):
# batch_idx = n_iter * opt['batch_size'] + b
# if batch_idx < len(dataset.inputs):
# video_name = dataset.inputs[batch_idx][0]
# output_cls[video_name].append(act_cls[b, :].detach().cpu().numpy())
# output_reg[video_name].append(act_reg[b, :].detach().cpu().numpy())
# if cls_label is not None:
# labels_cls[video_name].append(cls_label[b, :].cpu().numpy())
# if reg_label is not None:
# labels_reg[video_name].append(reg_label[b, :].cpu().numpy())
# except Exception as e:
# print(f"Error in batch {n_iter}: {str(e)}")
# continue
# # Stack arrays
# for video_name in dataset.video_list:
# if output_cls[video_name]:
# output_cls[video_name] = np.stack(output_cls[video_name], axis=0)
# output_reg[video_name] = np.stack(output_reg[video_name], axis=0)
# if labels_cls[video_name]:
# labels_cls[video_name] = np.stack(labels_cls[video_name], axis=0)
# if labels_reg[video_name]:
# labels_reg[video_name] = np.stack(labels_reg[video_name], axis=0)
# return output_cls, output_reg, labels_cls, labels_reg
# def eval_map_nms(opt, dataset, output_cls, output_reg):
# """Evaluate with Non-Maximum Suppression"""
# result_dict = {}
# anchors = opt['anchors']
# for video_name in dataset.video_list:
# if video_name not in output_cls or len(output_cls[video_name]) == 0:
# result_dict[video_name] = []
# continue
# duration = dataset.video_len[video_name]
# video_time = float(dataset.video_dict[video_name]["duration"])
# frame_to_time = 100.0 * video_time / duration
# proposal_dict = []
# for idx in range(min(duration, len(output_cls[video_name]))):
# cls_anc = output_cls[video_name][idx]
# reg_anc = output_reg[video_name][idx]
# for anc_idx in range(len(anchors)):
# if anc_idx >= len(cls_anc):
# continue
# cls = np.argwhere(cls_anc[anc_idx][:-1] > opt['threshold']).reshape(-1)
# if len(cls) == 0:
# continue
# ed = idx + anchors[anc_idx] * reg_anc[anc_idx][0]
# length = anchors[anc_idx] * np.exp(reg_anc[anc_idx][1])
# st = ed - length
# for cidx in range(len(cls)):
# label = cls[cidx]
# if label < len(dataset.label_name):
# tmp_dict = {
# "segment": [float(st * frame_to_time / 100.0), float(ed * frame_to_time / 100.0)],
# "score": float(cls_anc[anc_idx][label]),
# "label": dataset.label_name[label],
# "gentime": float(idx * frame_to_time / 100.0)
# }
# proposal_dict.append(tmp_dict)
# # Apply NMS
# proposal_dict = non_max_suppression(proposal_dict, overlapThresh=opt['soft_nms'])
# result_dict[video_name] = proposal_dict
# return result_dict
# def load_ground_truth(opt, video_name):
# """Load ground truth annotations if available"""
# gt_segments = []
# duration = 0
# try:
# video_anno_file = opt["video_anno"].format(opt["split"])
# if os.path.exists(video_anno_file):
# with open(video_anno_file, 'r') as f:
# anno_data = json.load(f)
# if video_name in anno_data['database']:
# gt_annotations = anno_data['database'][video_name]['annotations']
# duration = anno_data['database'][video_name]['duration']
# for anno in gt_annotations:
# start, end = anno['segment']
# gt_segments.append({
# 'label': anno['label'],
# 'start': start,
# 'end': end,
# 'duration': end - start
# })
# except Exception as e:
# print(f"Could not load ground truth: {str(e)}")
# return gt_segments, duration
# def process_video(video_name, split_number):
# """Process a single video for action localization"""
# try:
# # Parse options
# opt = opts.parse_opt()
# opt = vars(opt)
# opt['mode'] = 'test'
# opt['split'] = str(split_number)
# opt['checkpoint_path'] = './checkpoint'
# opt['video_feature_all_test'] = './data/I3D/'
# opt['anchors'] = [int(item) for item in opt['anchors'].split(',')]
# opt['batch_size'] = 1
# # Check if required files exist
# checkpoint_path = './checkpoint/01_ckp_best.pth.tar'
# if not os.path.exists(checkpoint_path):
# return "Error: Model checkpoint not found at ./checkpoint/01_ckp_best.pth.tar"
# npz_path = os.path.join(opt['video_feature_all_test'], f"{video_name}.npz")
# if not os.path.exists(npz_path):
# return f"Error: Feature file not found at {npz_path}"
# # Load model
# model = MYNET(opt).to(device)
# checkpoint = torch.load(checkpoint_path, map_location=device)
# # Handle different checkpoint formats
# if 'state_dict' in checkpoint:
# model.load_state_dict(checkpoint['state_dict'])
# else:
# model.load_state_dict(checkpoint)
# model.eval()
# # Create dataset
# dataset = VideoDataSet(opt, subset='test', video_name=video_name)
# if len(dataset.video_list) == 0:
# return f"Error: No video found with name '{video_name}' in dataset"
# # Run inference
# output_cls, output_reg, labels_cls, labels_reg = eval_frame(opt, model, dataset)
# result_dict = eval_map_nms(opt, dataset, output_cls, output_reg)
# # Load ground truth
# gt_segments, duration = load_ground_truth(opt, video_name)
# # Process predictions
# pred_segments = []
# for pred in result_dict.get(video_name, []):
# start, end = pred['segment']
# pred_segments.append({
# 'label': pred['label'],
# 'start': start,
# 'end': end,
# 'duration': end - start,
# 'score': pred['score']
# })
# # Generate output text
# output_text = f"Predicted Actions for Video: {video_name}\n"
# output_text += "=" * 50 + "\n\n"
# if pred_segments:
# output_text += "PREDICTED ACTIONS:\n"
# output_text += "-" * 30 + "\n"
# for i, pred in enumerate(pred_segments, 1):
# output_text += f"{i}. {pred['label']}\n"
# output_text += f" Time: [{pred['start']:.2f}s - {pred['end']:.2f}s]\n"
# output_text += f" Duration: {pred['duration']:.2f}s\n"
# output_text += f" Confidence: {pred['score']:.3f}\n\n"
# else:
# output_text += "No actions detected above threshold.\n\n"
# # Add ground truth comparison if available
# if gt_segments:
# output_text += "\nGROUND TRUTH COMPARISON:\n"
# output_text += "-" * 30 + "\n"
# # Calculate basic metrics
# matched_count = 0
# total_pred = len(pred_segments)
# total_gt = len(gt_segments)
# for gt in gt_segments:
# output_text += f"GT: {gt['label']} [{gt['start']:.2f}s - {gt['end']:.2f}s]\n"
# # Find best matching prediction
# best_match = None
# best_iou = 0
# for pred in pred_segments:
# # Simple overlap calculation
# overlap_start = max(gt['start'], pred['start'])
# overlap_end = min(gt['end'], pred['end'])
# if overlap_end > overlap_start:
# overlap = overlap_end - overlap_start
# union = (gt['end'] - gt['start']) + (pred['end'] - pred['start']) - overlap
# iou = overlap / union if union > 0 else 0
# if iou > best_iou:
# best_iou = iou
# best_match = pred
# if best_match and best_iou > VIS_CONFIG['iou_threshold']:
# matched_count += 1
# output_text += f" → Matched with: {best_match['label']} (IoU: {best_iou:.3f})\n"
# else:
# output_text += f" → No match found\n"
# output_text += "\n"
# # Summary statistics
# precision = matched_count / total_pred if total_pred > 0 else 0
# recall = matched_count / total_gt if total_gt > 0 else 0
# f1 = 2 * precision * recall / (precision + recall) if (precision + recall) > 0 else 0
# output_text += f"\nSUMMARY STATISTICS:\n"
# output_text += f"Total Predictions: {total_pred}\n"
# output_text += f"Total Ground Truth: {total_gt}\n"
# output_text += f"Matched: {matched_count}\n"
# output_text += f"Precision: {precision:.3f}\n"
# output_text += f"Recall: {recall:.3f}\n"
# output_text += f"F1-Score: {f1:.3f}\n"
# return output_text
# except Exception as e:
# return f"Error processing video: {str(e)}\n\nPlease check:\n1. Model checkpoint exists\n2. Feature file exists\n3. All dependencies are installed"
# def get_available_videos():
# """Get list of available videos from I3D features directory"""
# feature_dir = './data/I3D/'
# if not os.path.exists(feature_dir):
# return []
# videos = []
# for file in os.listdir(feature_dir):
# if file.endswith('.npz'):
# video_name = file.replace('.npz', '')
# videos.append(video_name)
# return sorted(videos)
# # Initialize available videos
# available_videos = get_available_videos()
# if not available_videos:
# available_videos = ["No videos found"]
# # Gradio Interface
# iface = gr.Interface(
# fn=process_video,
# inputs=[
# gr.Dropdown(
# label="Select Video",
# choices=available_videos,
# value=available_videos[0] if available_videos else None,
# info="Choose from pre-uploaded videos in data/I3D/ folder"
# ),
# gr.Dropdown(
# label="Split Number",
# choices=["1", "2", "3"],
# value="1",
# info="Dataset split for annotations"
# )
# ],
# outputs=[
# gr.Textbox(
# label="Action Predictions",
# lines=20,
# max_lines=50,
# show_copy_button=True
# )
# ],
# title="🎬 Temporal Action Localization",
# description="""
# This app performs temporal action localization on pre-uploaded videos using I3D features.
# **How to use:**
# 1. Select a video from the dropdown (videos must be in data/I3D/ folder as .npz files)
# 2. Choose the annotation split number
# 3. Click Submit to get action predictions
# **Requirements:**
# - Model checkpoint: `01_ckp_best.pth.tar` in root directory
# - Video features: `.npz` files in `data/I3D/` folder
# """,
# examples=[
# [available_videos[0] if available_videos and available_videos[0] != "No videos found" else "example_video", "1"],
# ] if available_videos and available_videos[0] != "No videos found" else None,
# cache_examples=False,
# theme=gr.themes.Soft()
# )
# if __name__ == '__main__':
# print(f"Available videos: {available_videos}")
# print(f"Using device: {device}")
# iface.launch(
# server_name="0.0.0.0",
# server_port=7860,
# share=False
# )
# import os
# import json
# import torch
# import numpy as np
# import gradio as gr
# import opts_egtea as opts
# from dataset import VideoDataSet, calc_iou
# from models import MYNET, SuppressNet
# from loss_func import cls_loss_func, regress_loss_func
# from eval import evaluation_detection
# from iou_utils import non_max_suppression, check_overlap_proposal
# from typing import List, Dict, Optional
# from huggingface_hub import hf_hub_download, list_repo_files
# import tempfile
# import shutil
# import traceback
# # Configuration
# VIS_CONFIG = {
# 'iou_threshold': 0.3,
# 'min_segment_duration': 1.0,
# }
# # Hugging Face Dataset Configuration
# HF_DATASET_REPO = "Darknsu/EGTEA_Dataset"
# HF_DATASET_SUBFOLDER = "I3D" # Adjust this based on your dataset structure
# # Determine device
# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# print(f"Using device: {device}")
# # Create local cache directory for downloaded files
# CACHE_DIR = "./hf_cache"
# os.makedirs(CACHE_DIR, exist_ok=True)
# def download_npz_file(video_name: str) -> str:
# """
# Download .npz file from Hugging Face dataset repository
# Returns: Local path to the downloaded file
# """
# try:
# # Construct the file path in the dataset repo
# file_path = f"{video_name}.npz"
# # Check if file already exists in cache
# local_path = os.path.join(CACHE_DIR, f"{video_name}.npz")
# if os.path.exists(local_path):
# print(f"Using cached file: {local_path}")
# return local_path
# # Download from Hugging Face dataset
# print(f"Downloading {file_path} from {HF_DATASET_REPO}...")
# downloaded_path = hf_hub_download(
# repo_id=HF_DATASET_REPO,
# filename=file_path,
# repo_type="dataset",
# cache_dir=CACHE_DIR
# )
# # Copy to our expected location for easier access
# shutil.copy2(downloaded_path, local_path)
# print(f"File downloaded and cached: {local_path}")
# return local_path
# except Exception as e:
# raise Exception(f"Failed to download {video_name}.npz: {str(e)}")
# def get_available_videos_from_hf():
# """Get list of available videos from Hugging Face dataset repository"""
# try:
# print("Fetching available videos from Hugging Face dataset...")
# files = list_repo_files(
# repo_id=HF_DATASET_REPO,
# repo_type="dataset"
# )
# # Filter for .npz files in the I3D subfolder
# videos = []
# for file in files:
# if file.startswith(f"") and file.endswith('.npz'):
# # Extract the full filename without extension
# # For files like "I3D/OP02-R02-TurkeySandwich.npz"
# video_name = os.path.basename(file).replace('.npz', '')
# videos.append(video_name)
# videos = sorted(videos)
# print(f"Found {len(videos)} videos in dataset: {videos[:5]}{'...' if len(videos) > 5 else ''}")
# return videos
# except Exception as e:
# print(f"Error fetching videos from HF dataset: {str(e)}")
# return ["Error loading videos"]
# class HFVideoDataSet(VideoDataSet):
# """
# Modified VideoDataSet that downloads files from Hugging Face on demand
# """
# def __init__(self, opt, subset='test', video_name=None):
# # Store the original video_feature_all_test path
# self.original_feature_path = opt['video_feature_all_test']
# # Create temporary directory for this session
# self.temp_dir = tempfile.mkdtemp(prefix="hf_video_")
# print(f"Created temp directory: {self.temp_dir}")
# # Download the specific video file if video_name is provided
# if video_name:
# try:
# print(f"Downloading features for video: {video_name}")
# downloaded_path = download_npz_file(video_name)
# # Ensure the temp directory exists
# os.makedirs(self.temp_dir, exist_ok=True)
# # Copy to temp directory with expected structure - FIX: Add proper path separator
# temp_file_path = os.path.join(self.temp_dir, f"{video_name}.npz")
# print(f"Copying {downloaded_path} to {temp_file_path}")
# shutil.copy2(downloaded_path, temp_file_path)
# # Verify file exists and print debug info
# if not os.path.exists(temp_file_path):
# raise Exception(f"Failed to copy file to {temp_file_path}")
# else:
# print(f"Video file ready: {temp_file_path}")
# print(f"File size: {os.path.getsize(temp_file_path)} bytes")
# except Exception as e:
# print(f"Error downloading video {video_name}: {str(e)}")
# # Clean up temp directory on error
# if hasattr(self, 'temp_dir') and os.path.exists(self.temp_dir):
# shutil.rmtree(self.temp_dir)
# raise e
# # Set the feature path to our temp directory
# opt['video_feature_all_test'] = self.temp_dir
# print(f"Set video_feature_all_test to: {opt['video_feature_all_test']}")
# # Initialize parent class
# try:
# super().__init__(opt, subset, video_name)
# print(f"Successfully initialized dataset with {len(self.video_list)} videos")
# except Exception as e:
# print(f"Error initializing parent VideoDataSet: {str(e)}")
# # Clean up temp directory on error
# if hasattr(self, 'temp_dir') and os.path.exists(self.temp_dir):
# shutil.rmtree(self.temp_dir)
# raise e
# def __del__(self):
# # Clean up temporary directory
# try:
# if hasattr(self, 'temp_dir') and os.path.exists(self.temp_dir):
# shutil.rmtree(self.temp_dir)
# print(f"Cleaned up temp directory: {self.temp_dir}")
# except Exception as e:
# print(f"Warning: Could not clean up temp directory: {e}")
# def eval_frame(opt, model, dataset):
# """Evaluate model frame by frame"""
# try:
# test_loader = torch.utils.data.DataLoader(
# dataset,
# batch_size=opt['batch_size'],
# shuffle=False,
# num_workers=0,
# pin_memory=False
# )
# labels_cls = {video_name: [] for video_name in dataset.video_list}
# labels_reg = {video_name: [] for video_name in dataset.video_list}
# output_cls = {video_name: [] for video_name in dataset.video_list}
# output_reg = {video_name: [] for video_name in dataset.video_list}
# model.eval()
# with torch.no_grad():
# for n_iter, batch_data in enumerate(test_loader):
# try:
# if len(batch_data) == 4:
# input_data, cls_label, reg_label, _ = batch_data
# else:
# input_data, cls_label, reg_label = batch_data
# input_data = input_data.to(device)
# cls_label = cls_label.to(device) if cls_label is not None else None
# reg_label = reg_label.to(device) if reg_label is not None else None
# act_cls, act_reg, _ = model(input_data.float())
# act_cls = torch.softmax(act_cls, dim=-1)
# for b in range(input_data.size(0)):
# batch_idx = n_iter * opt['batch_size'] + b
# if batch_idx < len(dataset.inputs):
# video_name = dataset.inputs[batch_idx][0]
# output_cls[video_name].append(act_cls[b, :].detach().cpu().numpy())
# output_reg[video_name].append(act_reg[b, :].detach().cpu().numpy())
# if cls_label is not None:
# labels_cls[video_name].append(cls_label[b, :].cpu().numpy())
# if reg_label is not None:
# labels_reg[video_name].append(reg_label[b, :].cpu().numpy())
# except Exception as e:
# print(f"Error in batch {n_iter}: {str(e)}")
# continue
# # Stack arrays
# for video_name in dataset.video_list:
# if output_cls[video_name]:
# output_cls[video_name] = np.stack(output_cls[video_name], axis=0)
# output_reg[video_name] = np.stack(output_reg[video_name], axis=0)
# if labels_cls[video_name]:
# labels_cls[video_name] = np.stack(labels_cls[video_name], axis=0)
# if labels_reg[video_name]:
# labels_reg[video_name] = np.stack(labels_reg[video_name], axis=0)
# return output_cls, output_reg, labels_cls, labels_reg
# except Exception as e:
# print(f"Error in eval_frame: {str(e)}")
# raise e
# def eval_map_nms(opt, dataset, output_cls, output_reg):
# """Evaluate with Non-Maximum Suppression"""
# try:
# result_dict = {}
# anchors = opt['anchors']
# for video_name in dataset.video_list:
# if video_name not in output_cls or len(output_cls[video_name]) == 0:
# result_dict[video_name] = []
# continue
# duration = dataset.video_len[video_name]
# video_time = float(dataset.video_dict[video_name]["duration"])
# frame_to_time = 100.0 * video_time / duration
# proposal_dict = []
# for idx in range(min(duration, len(output_cls[video_name]))):
# cls_anc = output_cls[video_name][idx]
# reg_anc = output_reg[video_name][idx]
# for anc_idx in range(len(anchors)):
# if anc_idx >= len(cls_anc):
# continue
# cls = np.argwhere(cls_anc[anc_idx][:-1] > opt['threshold']).reshape(-1)
# if len(cls) == 0:
# continue
# ed = idx + anchors[anc_idx] * reg_anc[anc_idx][0]
# length = anchors[anc_idx] * np.exp(reg_anc[anc_idx][1])
# st = ed - length
# for cidx in range(len(cls)):
# label = cls[cidx]
# if label < len(dataset.label_name):
# tmp_dict = {
# "segment": [float(st * frame_to_time / 100.0), float(ed * frame_to_time / 100.0)],
# "score": float(cls_anc[anc_idx][label]),
# "label": dataset.label_name[label],
# "gentime": float(idx * frame_to_time / 100.0)
# }
# proposal_dict.append(tmp_dict)
# # Apply NMS
# proposal_dict = non_max_suppression(proposal_dict, overlapThresh=opt['soft_nms'])
# result_dict[video_name] = proposal_dict
# return result_dict
# except Exception as e:
# print(f"Error in eval_map_nms: {str(e)}")
# raise e
# def load_ground_truth(opt, video_name):
# """Load ground truth annotations if available"""
# gt_segments = []
# duration = 0
# try:
# video_anno_file = opt["video_anno"].format(opt["split"])
# if os.path.exists(video_anno_file):
# with open(video_anno_file, 'r') as f:
# anno_data = json.load(f)
# if video_name in anno_data['database']:
# gt_annotations = anno_data['database'][video_name]['annotations']
# duration = anno_data['database'][video_name]['duration']
# for anno in gt_annotations:
# start, end = anno['segment']
# gt_segments.append({
# 'label': anno['label'],
# 'start': start,
# 'end': end,
# 'duration': end - start
# })
# except Exception as e:
# print(f"Could not load ground truth: {str(e)}")
# return gt_segments, duration
# def process_video(video_name, split_number, progress=gr.Progress()):
# """Process a single video for action localization"""
# dataset = None # Initialize dataset variable
# try:
# if not video_name or video_name in ["Error: Could not load videos from HF dataset", "Error loading videos"]:
# return "Error: Please select a valid video name"
# progress(0.1, desc="Initializing...")
# # Parse options
# opt = opts.parse_opt()
# opt = vars(opt)
# opt['mode'] = 'test'
# opt['split'] = str(split_number)
# opt['checkpoint_path'] = './checkpoint'
# opt['video_feature_all_test'] = './data/I3D/' # This will be overridden by HFVideoDataSet
# opt['anchors'] = [int(item) for item in opt['anchors'].split(',')]
# opt['batch_size'] = 1
# progress(0.2, desc="Checking model checkpoint...")
# # Check if required files exist
# checkpoint_path = './checkpoint/01_ckp_best.pth.tar'
# if not os.path.exists(checkpoint_path):
# # Try alternative locations
# alt_paths = ['./01_ckp_best.pth.tar', '01_ckp_best.pth.tar']
# checkpoint_path = None
# for alt_path in alt_paths:
# if os.path.exists(alt_path):
# checkpoint_path = alt_path
# break
# if checkpoint_path is None:
# return "Error: Model checkpoint not found. Please ensure '01_ckp_best.pth.tar' is in the repository."
# progress(0.3, desc="Loading model...")
# # Load model
# model = MYNET(opt).to(device)
# checkpoint = torch.load(checkpoint_path, map_location=device)
# # Handle different checkpoint formats
# if 'state_dict' in checkpoint:
# model.load_state_dict(checkpoint['state_dict'])
# else:
# model.load_state_dict(checkpoint)
# model.eval()
# print("Model loaded successfully")
# progress(0.4, desc=f"Downloading video features for {video_name}...")
# # Create dataset with HF integration
# try:
# dataset = HFVideoDataSet(opt, subset='test', video_name=video_name)
# print(f"Dataset created successfully with {len(dataset.video_list)} videos")
# except Exception as e:
# error_msg = f"Error downloading or loading video '{video_name}': {str(e)}\n\nPlease check:\n1. Video name is correct\n2. File exists in HF dataset\n3. Network connection is stable"
# print(error_msg)
# return error_msg
# if len(dataset.video_list) == 0:
# return f"Error: No video found with name '{video_name}' in dataset after download"
# progress(0.6, desc="Running inference...")
# # Run inference
# try:
# output_cls, output_reg, labels_cls, labels_reg = eval_frame(opt, model, dataset)
# print("Inference completed successfully")
# except Exception as e:
# error_msg = f"Error during inference: {str(e)}"
# print(error_msg)
# return error_msg
# progress(0.8, desc="Processing results...")
# try:
# result_dict = eval_map_nms(opt, dataset, output_cls, output_reg)
# print("NMS processing completed")
# except Exception as e:
# error_msg = f"Error during NMS processing: {str(e)}"
# print(error_msg)
# return error_msg
# # Load ground truth
# gt_segments, duration = load_ground_truth(opt, video_name)
# # Process predictions
# pred_segments = []
# for pred in result_dict.get(video_name, []):
# start, end = pred['segment']
# pred_segments.append({
# 'label': pred['label'],
# 'start': start,
# 'end': end,
# 'duration': end - start,
# 'score': pred['score']
# })
# progress(0.9, desc="Generating output...")
# # Generate output text
# output_text = f"Predicted Actions for Video: {video_name}\n"
# output_text += "=" * 50 + "\n\n"
# if pred_segments:
# output_text += "PREDICTED ACTIONS:\n"
# output_text += "-" * 30 + "\n"
# for i, pred in enumerate(pred_segments, 1):
# output_text += f"{i}. {pred['label']}\n"
# output_text += f" Time: [{pred['start']:.2f}s - {pred['end']:.2f}s]\n"
# output_text += f" Duration: {pred['duration']:.2f}s\n"
# output_text += f" Confidence: {pred['score']:.3f}\n\n"
# else:
# output_text += "No actions detected above threshold.\n\n"
# # Add ground truth comparison if available
# if gt_segments:
# output_text += "\nGROUND TRUTH COMPARISON:\n"
# output_text += "-" * 30 + "\n"
# # Calculate basic metrics
# matched_count = 0
# total_pred = len(pred_segments)
# total_gt = len(gt_segments)
# for gt in gt_segments:
# output_text += f"GT: {gt['label']} [{gt['start']:.2f}s - {gt['end']:.2f}s]\n"
# # Find best matching prediction
# best_match = None
# best_iou = 0
# for pred in pred_segments:
# # Simple overlap calculation
# overlap_start = max(gt['start'], pred['start'])
# overlap_end = min(gt['end'], pred['end'])
# if overlap_end > overlap_start:
# overlap = overlap_end - overlap_start
# union = (gt['end'] - gt['start']) + (pred['end'] - pred['start']) - overlap
# iou = overlap / union if union > 0 else 0
# if iou > best_iou:
# best_iou = iou
# best_match = pred
# if best_match and best_iou > VIS_CONFIG['iou_threshold']:
# matched_count += 1
# output_text += f" → Matched with: {best_match['label']} (IoU: {best_iou:.3f})\n"
# else:
# output_text += f" → No match found\n"
# output_text += "\n"
# # Summary statistics
# precision = matched_count / total_pred if total_pred > 0 else 0
# recall = matched_count / total_gt if total_gt > 0 else 0
# f1 = 2 * precision * recall / (precision + recall) if (precision + recall) > 0 else 0
# output_text += f"\nSUMMARY STATISTICS:\n"
# output_text += f"Total Predictions: {total_pred}\n"
# output_text += f"Total Ground Truth: {total_gt}\n"
# output_text += f"Matched: {matched_count}\n"
# output_text += f"Precision: {precision:.3f}\n"
# output_text += f"Recall: {recall:.3f}\n"
# output_text += f"F1-Score: {f1:.3f}\n"
# progress(1.0, desc="Complete!")
# print("Processing completed successfully")
# return output_text
# except Exception as e:
# error_details = traceback.format_exc()
# error_msg = f"Error processing video: {str(e)}\n\nDetailed error:\n{error_details}\n\nPlease check:\n1. Model checkpoint exists\n2. Video exists in HF dataset\n3. All dependencies are installed"
# print(error_msg)
# return error_msg
# finally:
# # Ensure cleanup happens even if there's an error
# if dataset is not None and hasattr(dataset, '__del__'):
# try:
# dataset.__del__()
# except Exception as e:
# print(f"Warning: Error during dataset cleanup: {e}")
# def refresh_video_list():
# """Refresh the list of available videos"""
# try:
# new_videos = get_available_videos_from_hf()
# return gr.Dropdown(choices=new_videos)
# except Exception as e:
# print(f"Error refreshing video list: {e}")
# return gr.Dropdown(choices=["Error refreshing videos"])
# # Initialize available videos
# print("Loading available videos from Hugging Face dataset...")
# try:
# available_videos = get_available_videos_from_hf()
# if not available_videos or available_videos == ["Error loading videos"]:
# available_videos = ["Error: Could not load videos from HF dataset"]
# except Exception as e:
# print(f"Error loading initial video list: {e}")
# available_videos = ["Error: Could not load videos from HF dataset"]
# print(f"Available videos: {len(available_videos)} videos found")
# # Gradio Interface
# with gr.Blocks(theme=gr.themes.Soft(), title="🎬 Temporal Action Localization") as iface:
# gr.Markdown("""
# # 🎬 Temporal Action Localization
# This app performs temporal action localization on videos using I3D features loaded dynamically from Hugging Face datasets.
# **Features:**
# - ✅ Dynamic loading from HF dataset repository
# - ✅ Real-time inference with progress tracking
# - ✅ Ground truth comparison when available
# - ✅ Detailed action predictions with confidence scores
# """)
# with gr.Row():
# with gr.Column(scale=1):
# video_dropdown = gr.Dropdown(
# label="Select Video",
# choices=available_videos,
# value=available_videos[0] if available_videos and "Error" not in available_videos[0] else None,
# info="Videos loaded from Hugging Face dataset"
# )
# split_dropdown = gr.Dropdown(
# label="Split Number",
# choices=["1", "2", "3"],
# value="1",
# info="Dataset split for annotations"
# )
# refresh_btn = gr.Button("🔄 Refresh Video List", variant="secondary")
# submit_btn = gr.Button("🚀 Run Action Localization", variant="primary")
# with gr.Column(scale=2):
# output_text = gr.Textbox(
# label="Action Predictions",
# lines=25,
# max_lines=50,
# show_copy_button=True,
# placeholder="Results will appear here..."
# )
# gr.Markdown(f"""
# **Dataset Source:** [{HF_DATASET_REPO}](https://huggingface.co/datasets/{HF_DATASET_REPO})
# **Requirements:**
# - Model checkpoint: `01_ckp_best.pth.tar` in repository root
# - Video features: Automatically downloaded from HF dataset
# """)
# # Event handlers
# refresh_btn.click(
# fn=refresh_video_list,
# outputs=video_dropdown
# )
# submit_btn.click(
# fn=process_video,
# inputs=[video_dropdown, split_dropdown],
# outputs=output_text
# )
# # Example
# if available_videos and "Error" not in available_videos[0]:
# gr.Examples(
# examples=[[available_videos[0], "1"]],
# inputs=[video_dropdown, split_dropdown],
# fn=process_video,
# outputs=output_text,
# cache_examples=False
# )
# if __name__ == '__main__':
# print(f"Available videos: {len(available_videos)}")
# print(f"Using device: {device}")
# print(f"HF Dataset: {HF_DATASET_REPO}")
# iface.launch(
# server_name="0.0.0.0",
# server_port=7860,
# share=False
# )
import os
import json
import torch
import numpy as np
import gradio as gr
import opts_egtea as opts
from dataset import VideoDataSet, calc_iou
from models import MYNET, SuppressNet
from loss_func import cls_loss_func, regress_loss_func
from eval import evaluation_detection
from iou_utils import non_max_suppression, check_overlap_proposal
from typing import List, Dict, Optional
from huggingface_hub import hf_hub_download, list_repo_files
import tempfile
import shutil
import traceback
# Configuration
VIS_CONFIG = {
'iou_threshold': 0.3,
'min_segment_duration': 1.0,
}
# Hugging Face Dataset Configuration
HF_DATASET_REPO = "Darknsu/EGTEA_Dataset"
HF_DATASET_SUBFOLDER = "I3D" # Adjust this based on your dataset structure
# Determine device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")
# Create local cache directory for downloaded files
CACHE_DIR = "./hf_cache"
os.makedirs(CACHE_DIR, exist_ok=True)
def download_npz_file(video_name: str) -> str:
"""
Download .npz file from Hugging Face dataset repository
Returns: Local path to the downloaded file
"""
try:
# Construct the file path in the dataset repo
file_path = f"{HF_DATASET_SUBFOLDER}/{video_name}.npz"
# Check if file already exists in cache
local_path = os.path.join(CACHE_DIR, f"{video_name}.npz")
if os.path.exists(local_path):
print(f"Using cached file: {local_path}")
return local_path
# Download from Hugging Face dataset
print(f"Downloading {file_path} from {HF_DATASET_REPO}...")
downloaded_path = hf_hub_download(
repo_id=HF_DATASET_REPO,
filename=file_path,
repo_type="dataset",
cache_dir=CACHE_DIR
)
# Copy to our expected location for easier access
shutil.copy2(downloaded_path, local_path)
print(f"File downloaded and cached: {local_path}")
return local_path
except Exception as e:
raise Exception(f"Failed to download {video_name}.npz: {str(e)}")
def get_available_videos_from_hf():
"""Get list of available videos from Hugging Face dataset repository"""
try:
print("Fetching available videos from Hugging Face dataset...")
files = list_repo_files(
repo_id=HF_DATASET_REPO,
repo_type="dataset"
)
# Filter for .npz files in the I3D subfolder
videos = []
for file in files:
if file.startswith(f"{HF_DATASET_SUBFOLDER}/") and file.endswith('.npz'):
# Extract the full filename without extension
# For files like "I3D/OP02-R02-TurkeySandwich.npz"
video_name = os.path.basename(file).replace('.npz', '')
videos.append(video_name)
videos = sorted(videos)
print(f"Found {len(videos)} videos in dataset: {videos[:5]}{'...' if len(videos) > 5 else ''}")
return videos
except Exception as e:
print(f"Error fetching videos from HF dataset: {str(e)}")
return ["Error loading videos"]
class HFVideoDataSet(VideoDataSet):
"""
Modified VideoDataSet that downloads files from Hugging Face on demand
"""
def __init__(self, opt, subset='test', video_name=None):
# Store the original video_feature_all_test path
self.original_feature_path = opt['video_feature_all_test']
# Create temporary directory for this session
self.temp_dir = tempfile.mkdtemp(prefix="hf_video_")
print(f"Created temp directory: {self.temp_dir}")
# Download the specific video file if video_name is provided
if video_name:
try:
print(f"Downloading features for video: {video_name}")
downloaded_path = download_npz_file(video_name)
# Ensure the temp directory exists
os.makedirs(self.temp_dir, exist_ok=True)
# Copy to temp directory with expected structure - FIX: Add proper path separator
temp_file_path = os.path.join(self.temp_dir, f"{video_name}.npz")
print(f"Copying {downloaded_path} to {temp_file_path}")
shutil.copy2(downloaded_path, temp_file_path)
# Verify file exists and print debug info
if not os.path.exists(temp_file_path):
raise Exception(f"Failed to copy file to {temp_file_path}")
else:
print(f"Video file ready: {temp_file_path}")
print(f"File size: {os.path.getsize(temp_file_path)} bytes")
except Exception as e:
print(f"Error downloading video {video_name}: {str(e)}")
# Clean up temp directory on error
if hasattr(self, 'temp_dir') and os.path.exists(self.temp_dir):
shutil.rmtree(self.temp_dir)
raise e
# Set the feature path to our temp directory
opt['video_feature_all_test'] = self.temp_dir
print(f"Set video_feature_all_test to: {opt['video_feature_all_test']}")
# Initialize parent class
try:
super().__init__(opt, subset, video_name)
print(f"Successfully initialized dataset with {len(self.video_list)} videos")
except Exception as e:
print(f"Error initializing parent VideoDataSet: {str(e)}")
# Clean up temp directory on error
if hasattr(self, 'temp_dir') and os.path.exists(self.temp_dir):
shutil.rmtree(self.temp_dir)
raise e
def __del__(self):
# Clean up temporary directory
try:
if hasattr(self, 'temp_dir') and os.path.exists(self.temp_dir):
shutil.rmtree(self.temp_dir)
print(f"Cleaned up temp directory: {self.temp_dir}")
except Exception as e:
print(f"Warning: Could not clean up temp directory: {e}")
def eval_frame(opt, model, dataset):
"""Evaluate model frame by frame"""
try:
test_loader = torch.utils.data.DataLoader(
dataset,
batch_size=opt['batch_size'],
shuffle=False,
num_workers=0,
pin_memory=False
)
labels_cls = {video_name: [] for video_name in dataset.video_list}
labels_reg = {video_name: [] for video_name in dataset.video_list}
output_cls = {video_name: [] for video_name in dataset.video_list}
output_reg = {video_name: [] for video_name in dataset.video_list}
model.eval()
with torch.no_grad():
for n_iter, batch_data in enumerate(test_loader):
try:
if len(batch_data) == 4:
input_data, cls_label, reg_label, _ = batch_data
else:
input_data, cls_label, reg_label = batch_data
input_data = input_data.to(device)
cls_label = cls_label.to(device) if cls_label is not None else None
reg_label = reg_label.to(device) if reg_label is not None else None
act_cls, act_reg, _ = model(input_data.float())
act_cls = torch.softmax(act_cls, dim=-1)
for b in range(input_data.size(0)):
batch_idx = n_iter * opt['batch_size'] + b
if batch_idx < len(dataset.inputs):
video_name = dataset.inputs[batch_idx][0]
output_cls[video_name].append(act_cls[b, :].detach().cpu().numpy())
output_reg[video_name].append(act_reg[b, :].detach().cpu().numpy())
if cls_label is not None:
labels_cls[video_name].append(cls_label[b, :].cpu().numpy())
if reg_label is not None:
labels_reg[video_name].append(reg_label[b, :].cpu().numpy())
except Exception as e:
print(f"Error in batch {n_iter}: {str(e)}")
continue
# Stack arrays
for video_name in dataset.video_list:
if output_cls[video_name]:
output_cls[video_name] = np.stack(output_cls[video_name], axis=0)
output_reg[video_name] = np.stack(output_reg[video_name], axis=0)
if labels_cls[video_name]:
labels_cls[video_name] = np.stack(labels_cls[video_name], axis=0)
if labels_reg[video_name]:
labels_reg[video_name] = np.stack(labels_reg[video_name], axis=0)
return output_cls, output_reg, labels_cls, labels_reg
except Exception as e:
print(f"Error in eval_frame: {str(e)}")
raise e
def eval_map_nms(opt, dataset, output_cls, output_reg):
"""Evaluate with Non-Maximum Suppression"""
try:
result_dict = {}
anchors = opt['anchors']
for video_name in dataset.video_list:
if video_name not in output_cls or len(output_cls[video_name]) == 0:
result_dict[video_name] = []
continue
duration = dataset.video_len[video_name]
video_time = float(dataset.video_dict[video_name]["duration"])
frame_to_time = 100.0 * video_time / duration
proposal_dict = []
for idx in range(min(duration, len(output_cls[video_name]))):
cls_anc = output_cls[video_name][idx]
reg_anc = output_reg[video_name][idx]
for anc_idx in range(len(anchors)):
if anc_idx >= len(cls_anc):
continue
cls = np.argwhere(cls_anc[anc_idx][:-1] > opt['threshold']).reshape(-1)
if len(cls) == 0:
continue
ed = idx + anchors[anc_idx] * reg_anc[anc_idx][0]
length = anchors[anc_idx] * np.exp(reg_anc[anc_idx][1])
st = ed - length
for cidx in range(len(cls)):
label = cls[cidx]
if label < len(dataset.label_name):
tmp_dict = {
"segment": [float(st * frame_to_time / 100.0), float(ed * frame_to_time / 100.0)],
"score": float(cls_anc[anc_idx][label]),
"label": dataset.label_name[label],
"gentime": float(idx * frame_to_time / 100.0)
}
proposal_dict.append(tmp_dict)
# Apply NMS
proposal_dict = non_max_suppression(proposal_dict, overlapThresh=opt['soft_nms'])
result_dict[video_name] = proposal_dict
return result_dict
except Exception as e:
print(f"Error in eval_map_nms: {str(e)}")
raise e
def load_ground_truth(opt, video_name):
"""Load ground truth annotations if available"""
gt_segments = []
duration = 0
try:
video_anno_file = opt["video_anno"].format(opt["split"])
if os.path.exists(video_anno_file):
with open(video_anno_file, 'r') as f:
anno_data = json.load(f)
if video_name in anno_data['database']:
gt_annotations = anno_data['database'][video_name]['annotations']
duration = anno_data['database'][video_name]['duration']
for anno in gt_annotations:
start, end = anno['segment']
gt_segments.append({
'label': anno['label'],
'start': start,
'end': end,
'duration': end - start
})
except Exception as e:
print(f"Could not load ground truth: {str(e)}")
return gt_segments, duration
def process_video(video_name, split_number, progress=gr.Progress()):
"""Process a single video for action localization"""
dataset = None # Initialize dataset variable
try:
if not video_name or video_name in ["Error: Could not load videos from HF dataset", "Error loading videos"]:
return "Error: Please select a valid video name"
progress(0.1, desc="Initializing...")
# Parse options
opt = opts.parse_opt()
opt = vars(opt)
opt['mode'] = 'test'
opt['split'] = str(split_number)
opt['checkpoint_path'] = './checkpoint'
opt['video_feature_all_test'] = './data/I3D/' # This will be overridden by HFVideoDataSet
opt['anchors'] = [int(item) for item in opt['anchors'].split(',')]
opt['batch_size'] = 1
progress(0.2, desc="Checking model checkpoint...")
# Check if required files exist
checkpoint_path = './checkpoint/01_ckp_best.pth.tar'
if not os.path.exists(checkpoint_path):
# Try alternative locations
alt_paths = ['./01_ckp_best.pth.tar', '01_ckp_best.pth.tar']
checkpoint_path = None
for alt_path in alt_paths:
if os.path.exists(alt_path):
checkpoint_path = alt_path
break
if checkpoint_path is None:
return "Error: Model checkpoint not found. Please ensure '01_ckp_best.pth.tar' is in the repository."
progress(0.3, desc="Loading model...")
# Load model
model = MYNET(opt).to(device)
checkpoint = torch.load(checkpoint_path, map_location=device)
# Handle different checkpoint formats
if 'state_dict' in checkpoint:
model.load_state_dict(checkpoint['state_dict'])
else:
model.load_state_dict(checkpoint)
model.eval()
print("Model loaded successfully")
progress(0.4, desc=f"Downloading video features for {video_name}...")
# Create dataset with HF integration
try:
dataset = HFVideoDataSet(opt, subset='test', video_name=video_name)
print(f"Dataset created successfully with {len(dataset.video_list)} videos")
except Exception as e:
error_msg = f"Error downloading or loading video '{video_name}': {str(e)}\n\nPlease check:\n1. Video name is correct\n2. File exists in HF dataset\n3. Network connection is stable"
print(error_msg)
return error_msg
if len(dataset.video_list) == 0:
return f"Error: No video found with name '{video_name}' in dataset after download"
progress(0.6, desc="Running inference...")
# Run inference
try:
output_cls, output_reg, labels_cls, labels_reg = eval_frame(opt, model, dataset)
print("Inference completed successfully")
except Exception as e:
error_msg = f"Error during inference: {str(e)}"
print(error_msg)
return error_msg
progress(0.8, desc="Processing results...")
try:
result_dict = eval_map_nms(opt, dataset, output_cls, output_reg)
print("NMS processing completed")
except Exception as e:
error_msg = f"Error during NMS processing: {str(e)}"
print(error_msg)
return error_msg
# Load ground truth
gt_segments, duration = load_ground_truth(opt, video_name)
# Process predictions
pred_segments = []
for pred in result_dict.get(video_name, []):
start, end = pred['segment']
pred_segments.append({
'label': pred['label'],
'start': start,
'end': end,
'duration': end - start,
'score': pred['score']
})
progress(0.9, desc="Generating output...")
# Generate output text
output_text = f"Predicted Actions for Video: {video_name}\n"
output_text += "=" * 50 + "\n\n"
if pred_segments:
output_text += "PREDICTED ACTIONS:\n"
output_text += "-" * 30 + "\n"
for i, pred in enumerate(pred_segments, 1):
output_text += f"{i}. {pred['label']}\n"
output_text += f" Time: [{pred['start']:.2f}s - {pred['end']:.2f}s]\n"
output_text += f" Duration: {pred['duration']:.2f}s\n"
output_text += f" Confidence: {pred['score']:.3f}\n\n"
else:
output_text += "No actions detected above threshold.\n\n"
# Add ground truth comparison if available
if gt_segments:
output_text += "\nGROUND TRUTH COMPARISON:\n"
output_text += "-" * 30 + "\n"
# Calculate basic metrics
matched_count = 0
total_pred = len(pred_segments)
total_gt = len(gt_segments)
for gt in gt_segments:
output_text += f"GT: {gt['label']} [{gt['start']:.2f}s - {gt['end']:.2f}s]\n"
# Find best matching prediction
best_match = None
best_iou = 0
for pred in pred_segments:
# Simple overlap calculation
overlap_start = max(gt['start'], pred['start'])
overlap_end = min(gt['end'], pred['end'])
if overlap_end > overlap_start:
overlap = overlap_end - overlap_start
union = (gt['end'] - gt['start']) + (pred['end'] - pred['start']) - overlap
iou = overlap / union if union > 0 else 0
if iou > best_iou:
best_iou = iou
best_match = pred
if best_match and best_iou > VIS_CONFIG['iou_threshold']:
matched_count += 1
output_text += f" → Matched with: {best_match['label']} (IoU: {best_iou:.3f})\n"
else:
output_text += f" → No match found\n"
output_text += "\n"
# Summary statistics
precision = matched_count / total_pred if total_pred > 0 else 0
recall = matched_count / total_gt if total_gt > 0 else 0
f1 = 2 * precision * recall / (precision + recall) if (precision + recall) > 0 else 0
output_text += f"\nSUMMARY STATISTICS:\n"
output_text += f"Total Predictions: {total_pred}\n"
output_text += f"Total Ground Truth: {total_gt}\n"
output_text += f"Matched: {matched_count}\n"
output_text += f"Precision: {precision:.3f}\n"
output_text += f"Recall: {recall:.3f}\n"
output_text += f"F1-Score: {f1:.3f}\n"
progress(1.0, desc="Complete!")
print("Processing completed successfully")
return output_text
except Exception as e:
error_details = traceback.format_exc()
error_msg = f"Error processing video: {str(e)}\n\nDetailed error:\n{error_details}\n\nPlease check:\n1. Model checkpoint exists\n2. Video exists in HF dataset\n3. All dependencies are installed"
print(error_msg)
return error_msg
finally:
# Ensure cleanup happens even if there's an error
if dataset is not None and hasattr(dataset, '__del__'):
try:
dataset.__del__()
except Exception as e:
print(f"Warning: Error during dataset cleanup: {e}")
def refresh_video_list():
"""Refresh the list of available videos"""
try:
new_videos = get_available_videos_from_hf()
return gr.Dropdown(choices=new_videos)
except Exception as e:
print(f"Error refreshing video list: {e}")
return gr.Dropdown(choices=["Error refreshing videos"])
# Initialize available videos
print("Loading available videos from Hugging Face dataset...")
try:
available_videos = get_available_videos_from_hf()
if not available_videos or available_videos == ["Error loading videos"]:
available_videos = ["Error: Could not load videos from HF dataset"]
except Exception as e:
print(f"Error loading initial video list: {e}")
available_videos = ["Error: Could not load videos from HF dataset"]
print(f"Available videos: {len(available_videos)} videos found")
# Gradio Interface
with gr.Blocks(theme=gr.themes.Soft(), title="🎬 Temporal Action Localization") as iface:
gr.Markdown("""
# 🎬 Temporal Action Localization
This app performs temporal action localization on videos using I3D features loaded dynamically from Hugging Face datasets.
**Features:**
- ✅ Dynamic loading from HF dataset repository
- ✅ Real-time inference with progress tracking
- ✅ Ground truth comparison when available
- ✅ Detailed action predictions with confidence scores
""")
with gr.Row():
with gr.Column(scale=1):
video_dropdown = gr.Dropdown(
label="Select Video",
choices=available_videos,
value=available_videos[0] if available_videos and "Error" not in available_videos[0] else None,
info="Videos loaded from Hugging Face dataset"
)
split_dropdown = gr.Dropdown(
label="Split Number",
choices=["1", "2", "3"],
value="1",
info="Dataset split for annotations"
)
refresh_btn = gr.Button("🔄 Refresh Video List", variant="secondary")
submit_btn = gr.Button("🚀 Run Action Localization", variant="primary")
with gr.Column(scale=2):
output_text = gr.Textbox(
label="Action Predictions",
lines=25,
max_lines=50,
show_copy_button=True,
placeholder="Results will appear here..."
)
gr.Markdown(f"""
**Dataset Source:** [{HF_DATASET_REPO}](https://huggingface.co/datasets/{HF_DATASET_REPO})
**Requirements:**
- Model checkpoint: `01_ckp_best.pth.tar` in repository root
- Video features: Automatically downloaded from HF dataset
""")
# Event handlers
refresh_btn.click(
fn=refresh_video_list,
outputs=video_dropdown
)
submit_btn.click(
fn=process_video,
inputs=[video_dropdown, split_dropdown],
outputs=output_text
)
# Example
if available_videos and "Error" not in available_videos[0]:
gr.Examples(
examples=[[available_videos[0], "1"]],
inputs=[video_dropdown, split_dropdown],
fn=process_video,
outputs=output_text,
cache_examples=False
)
if __name__ == '__main__':
print(f"Available videos: {len(available_videos)}")
print(f"Using device: {device}")
print(f"HF Dataset: {HF_DATASET_REPO}")
iface.launch(
server_name="0.0.0.0",
server_port=7860,
share=False
) |