Open15
関節アノテーション 肩・肘・膝
PINTO831
1005
-
000000013455.jpg - 620
-
000000003786.jpg - 169
-
000000010104.jpg - 453
PINTOpython train_dual.py \
--workers 8 \
--device 0 \
--batch 16 \
--data data/original.yaml \
--img 640 \
--cfg models/detect/yolov9-t_original.yaml \
--weights best-t.pt \
--name yolov9-shoulder-elbow-knee-t \
--hyp hyp.scratch-high_original.yaml \
--min-items 0 \
--epochs 100 \
--close-mosaic 15
- 3x3ピクセル, 727枚, T, 肩肘膝3クラスのみ
Class Images Instances P R mAP50 mAP50-95
all 146 2868 0.0908 0.0316 0.0171 0.00537
shoulder 146 1296 0.117 0.0563 0.0259 0.00716
elbow 146 877 0.0988 0.0182 0.0139 0.00492
knee 146 695 0.0569 0.0201 0.0114 0.00402
names:
0: body
1: adult
2: child
3: male
4: female
5: body_with_wheelchair
6: body_with_crutches
7: head
8: front
9: right-front
10: right-side
11: right-back
12: back
13: left-back
14: left-side
15: left-front
16: face
17: eye
18: nose
19: mouth
20: ear
21: hand
22: hand_left
23: hand_right
24: foot
25: shoulder
26: elbow
27: knee
python train_dual.py \
--workers 8 \
--device 0 \
--batch 2 \
--data data/original.yaml \
--img 640 \
--cfg models/detect/yolov9-t_original.yaml \
--weights best-t.pt \
--name yolov9-shoulder-elbow-knee-t \
--hyp hyp.scratch-high_original.yaml \
--min-items 0 \
--epochs 100 \
--close-mosaic 15
- 3x3ピクセル, 727枚, T, 28クラス
Class Images Instances P R mAP50 mAP50-95
all 146 10714 0.54 0.4 0.412 0.275
body 146 1156 0.683 0.64 0.676 0.492
adult 146 761 0.762 0.692 0.723 0.551
child 146 80 0.331 0.35 0.375 0.307
male 146 512 0.638 0.68 0.661 0.512
female 146 247 0.515 0.615 0.557 0.449
head 146 847 0.734 0.791 0.827 0.56
front 146 138 0.47 0.413 0.42 0.332
right-front 146 190 0.505 0.305 0.325 0.263
right-side 146 128 0.459 0.328 0.33 0.261
right-back 146 66 0.453 0.301 0.306 0.238
back 146 66 0.223 0.212 0.125 0.0932
left-back 146 36 0.176 0.167 0.154 0.115
left-side 146 83 0.467 0.337 0.308 0.226
left-front 146 140 0.472 0.28 0.26 0.214
face 146 396 0.827 0.712 0.757 0.536
eye 146 374 0.698 0.265 0.323 0.111
nose 146 335 0.692 0.427 0.464 0.205
mouth 146 278 0.557 0.342 0.33 0.13
ear 146 320 0.679 0.403 0.462 0.234
hand 146 579 0.837 0.615 0.705 0.408
hand_left 146 291 0.693 0.536 0.594 0.344
hand_right 146 288 0.591 0.465 0.504 0.306
foot 146 535 0.568 0.521 0.526 0.274
shoulder 146 1296 1 0 0 0
elbow 146 877 0 0 0 0
knee 146 695 0 0 0 0
- 6x6ピクセル, 727枚, T, 28クラス
Class Images Instances P R mAP50 mAP50-95
all 146 10714 0.519 0.393 0.417 0.28
body 146 1156 0.655 0.639 0.663 0.479
adult 146 761 0.809 0.628 0.693 0.537
child 146 80 0.295 0.388 0.357 0.304
male 146 512 0.619 0.664 0.661 0.518
female 146 247 0.575 0.597 0.602 0.485
head 146 847 0.779 0.797 0.834 0.564
front 146 138 0.454 0.435 0.435 0.348
right-front 146 190 0.503 0.311 0.345 0.277
right-side 146 128 0.47 0.32 0.34 0.263
right-back 146 66 0.433 0.273 0.306 0.24
back 146 66 0.272 0.203 0.132 0.0964
left-back 146 36 0.3 0.194 0.193 0.148
left-side 146 83 0.528 0.313 0.306 0.23
left-front 146 140 0.463 0.277 0.282 0.225
face 146 396 0.83 0.712 0.763 0.549
eye 146 374 0.578 0.238 0.287 0.101
nose 146 335 0.684 0.418 0.472 0.207
mouth 146 278 0.628 0.324 0.362 0.141
ear 146 320 0.697 0.394 0.445 0.216
hand 146 579 0.855 0.589 0.694 0.408
hand_left 146 291 0.676 0.54 0.59 0.339
hand_right 146 288 0.602 0.435 0.504 0.309
foot 146 535 0.568 0.518 0.524 0.288
shoulder 146 1296 0.216 0.0201 0.0261 0.00512
elbow 146 877 0 0 0.00907 0.00182
knee 146 695 0 0 0.0148 0.00274
- 9x9ピクセル, 727枚, T, 28クラス
Class Images Instances P R mAP50 mAP50-95
all 146 10714 0.519 0.409 0.422 0.283
body 146 1156 0.689 0.63 0.675 0.493
adult 146 761 0.783 0.619 0.675 0.533
child 146 80 0.375 0.412 0.386 0.349
male 146 512 0.624 0.635 0.626 0.504
female 146 247 0.54 0.652 0.593 0.479
head 146 847 0.72 0.804 0.835 0.571
front 146 138 0.483 0.449 0.429 0.338
right-front 146 190 0.503 0.295 0.357 0.285
right-side 146 128 0.434 0.32 0.342 0.264
right-back 146 66 0.429 0.284 0.304 0.239
back 146 66 0.303 0.227 0.155 0.114
left-back 146 36 0.243 0.222 0.165 0.126
left-side 146 83 0.435 0.361 0.308 0.234
left-front 146 140 0.451 0.258 0.279 0.232
face 146 396 0.794 0.725 0.781 0.551
eye 146 374 0.568 0.246 0.296 0.106
nose 146 335 0.678 0.442 0.49 0.217
mouth 146 278 0.529 0.295 0.317 0.136
ear 146 320 0.616 0.391 0.455 0.222
hand 146 579 0.814 0.606 0.686 0.398
hand_left 146 291 0.634 0.546 0.569 0.326
hand_right 146 288 0.574 0.458 0.49 0.296
foot 146 535 0.523 0.525 0.528 0.288
shoulder 146 1296 0.356 0.143 0.141 0.0382
elbow 146 877 0.199 0.0342 0.0462 0.012
knee 146 695 0.203 0.0417 0.0539 0.0138
- 12x12ピクセル, 727枚, T, 28クラス
Class Images Instances P R mAP50 mAP50-95
all 146 10714 0.553 0.41 0.435 0.287
body 146 1156 0.757 0.619 0.687 0.5
adult 146 761 0.795 0.591 0.652 0.516
child 146 80 0.404 0.35 0.351 0.291
male 146 512 0.663 0.645 0.656 0.513
female 146 247 0.524 0.591 0.553 0.451
head 146 847 0.752 0.786 0.83 0.573
front 146 138 0.524 0.435 0.434 0.35
right-front 146 190 0.613 0.284 0.364 0.295
right-side 146 128 0.45 0.328 0.345 0.27
right-back 146 66 0.459 0.273 0.313 0.246
back 146 66 0.269 0.197 0.136 0.101
left-back 146 36 0.26 0.25 0.214 0.16
left-side 146 83 0.457 0.337 0.325 0.253
left-front 146 140 0.423 0.279 0.277 0.231
face 146 396 0.835 0.691 0.761 0.556
eye 146 374 0.563 0.251 0.302 0.111
nose 146 335 0.699 0.416 0.488 0.216
mouth 146 278 0.553 0.324 0.339 0.136
ear 146 320 0.721 0.409 0.471 0.226
hand 146 579 0.834 0.604 0.695 0.401
hand_left 146 291 0.665 0.525 0.585 0.331
hand_right 146 288 0.591 0.476 0.52 0.316
foot 146 535 0.546 0.525 0.536 0.289
shoulder 146 1296 0.412 0.292 0.259 0.0787
elbow 146 877 0.307 0.0889 0.0984 0.0267
knee 146 695 0.312 0.104 0.117 0.033
- 15x15ピクセル, 727枚, T, 28クラス
Class Images Instances P R mAP50 mAP50-95
all 146 10714 0.55 0.411 0.438 0.289
body 146 1156 0.781 0.563 0.653 0.485
adult 146 761 0.793 0.567 0.658 0.527
child 146 80 0.444 0.388 0.395 0.351
male 146 512 0.667 0.623 0.649 0.514
female 146 247 0.537 0.573 0.564 0.466
head 146 847 0.768 0.79 0.838 0.568
front 146 138 0.51 0.435 0.433 0.351
right-front 146 190 0.508 0.3 0.365 0.293
right-side 146 128 0.455 0.293 0.353 0.275
right-back 146 66 0.378 0.242 0.283 0.226
back 146 66 0.258 0.158 0.158 0.113
left-back 146 36 0.226 0.195 0.141 0.107
left-side 146 83 0.471 0.301 0.311 0.235
left-front 146 140 0.488 0.279 0.286 0.234
face 146 396 0.798 0.705 0.77 0.559
eye 146 374 0.604 0.241 0.309 0.11
nose 146 335 0.699 0.415 0.488 0.218
mouth 146 278 0.572 0.342 0.364 0.15
ear 146 320 0.661 0.378 0.442 0.217
hand 146 579 0.824 0.582 0.67 0.392
hand_left 146 291 0.672 0.508 0.562 0.322
hand_right 146 288 0.575 0.441 0.492 0.297
foot 146 535 0.582 0.54 0.539 0.296
shoulder 146 1296 0.349 0.393 0.297 0.0949
elbow 146 877 0.319 0.195 0.161 0.0493
knee 146 695 0.369 0.245 0.207 0.0635
- 18x18ピクセル, 727枚, T, 28クラス
Class Images Instances P R mAP50 mAP50-95
all 146 10714 0.556 0.416 0.438 0.288
body 146 1156 0.829 0.542 0.654 0.484
adult 146 761 0.776 0.564 0.655 0.515
child 146 80 0.396 0.35 0.377 0.316
male 146 512 0.656 0.586 0.623 0.49
female 146 247 0.512 0.567 0.537 0.428
head 146 847 0.789 0.777 0.831 0.576
front 146 138 0.497 0.442 0.446 0.362
right-front 146 190 0.534 0.311 0.36 0.289
right-side 146 128 0.465 0.305 0.344 0.278
right-back 146 66 0.418 0.258 0.304 0.239
back 146 66 0.279 0.199 0.154 0.104
left-back 146 36 0.214 0.194 0.123 0.0856
left-side 146 83 0.518 0.313 0.326 0.25
left-front 146 140 0.446 0.264 0.28 0.229
face 146 396 0.804 0.693 0.759 0.549
eye 146 374 0.576 0.247 0.299 0.11
nose 146 335 0.677 0.4 0.468 0.214
mouth 146 278 0.594 0.331 0.364 0.147
ear 146 320 0.666 0.356 0.433 0.211
hand 146 579 0.839 0.579 0.667 0.396
hand_left 146 291 0.663 0.488 0.557 0.332
hand_right 146 288 0.602 0.435 0.506 0.317
foot 146 535 0.595 0.525 0.525 0.285
shoulder 146 1296 0.373 0.485 0.302 0.103
elbow 146 877 0.358 0.259 0.206 0.0691
knee 146 695 0.371 0.34 0.281 0.0971
- 21x21ピクセル, 727枚, T, 28クラス
Class Images Instances P R mAP50 mAP50-95
all 146 10714 0.578 0.41 0.448 0.292
body 146 1156 0.885 0.525 0.652 0.487
adult 146 761 0.825 0.532 0.644 0.515
child 146 80 0.407 0.4 0.353 0.301
male 146 512 0.679 0.578 0.628 0.503
female 146 247 0.587 0.494 0.544 0.451
head 146 847 0.84 0.764 0.826 0.569
front 146 138 0.493 0.435 0.451 0.362
right-front 146 190 0.538 0.3 0.368 0.3
right-side 146 128 0.535 0.297 0.361 0.278
right-back 146 66 0.458 0.258 0.296 0.232
back 146 66 0.255 0.167 0.136 0.0994
left-back 146 36 0.176 0.167 0.126 0.0878
left-side 146 83 0.487 0.301 0.338 0.247
left-front 146 140 0.474 0.25 0.28 0.23
face 146 396 0.849 0.694 0.768 0.557
eye 146 374 0.59 0.227 0.295 0.106
nose 146 335 0.672 0.391 0.462 0.204
mouth 146 278 0.617 0.336 0.374 0.148
ear 146 320 0.72 0.362 0.458 0.229
hand 146 579 0.869 0.563 0.673 0.398
hand_left 146 291 0.7 0.465 0.56 0.325
hand_right 146 288 0.603 0.417 0.483 0.296
foot 146 535 0.597 0.516 0.534 0.289
shoulder 146 1296 0.416 0.556 0.434 0.164
elbow 146 877 0.38 0.301 0.256 0.095
knee 146 695 0.391 0.377 0.337 0.129
- 30x30ピクセル, 727枚, T, 28クラス
Class Images Instances P R mAP50 mAP50-95
all 146 10714 0.572 0.414 0.453 0.294
body 146 1156 0.846 0.51 0.635 0.476
adult 146 761 0.796 0.499 0.615 0.496
child 146 80 0.378 0.375 0.361 0.314
male 146 512 0.652 0.564 0.611 0.494
female 146 247 0.533 0.543 0.552 0.444
head 146 847 0.84 0.751 0.816 0.561
front 146 138 0.472 0.428 0.444 0.361
right-front 146 190 0.546 0.274 0.348 0.281
right-side 146 128 0.45 0.289 0.347 0.273
right-back 146 66 0.42 0.258 0.291 0.226
back 146 66 0.307 0.136 0.142 0.117
left-back 146 36 0.279 0.167 0.173 0.136
left-side 146 83 0.473 0.289 0.302 0.222
left-front 146 140 0.392 0.243 0.257 0.214
face 146 396 0.812 0.679 0.756 0.558
eye 146 374 0.572 0.232 0.294 0.105
nose 146 335 0.645 0.409 0.465 0.212
mouth 146 278 0.616 0.299 0.355 0.141
ear 146 320 0.733 0.366 0.435 0.217
hand 146 579 0.877 0.546 0.668 0.39
hand_left 146 291 0.701 0.43 0.555 0.322
hand_right 146 288 0.604 0.441 0.485 0.295
foot 146 535 0.648 0.492 0.517 0.282
shoulder 146 1296 0.45 0.658 0.569 0.219
elbow 146 877 0.392 0.395 0.342 0.125
knee 146 695 0.427 0.499 0.437 0.162
PINTOデータ調整後
python train_dual.py \
--workers 8 \
--device 0 \
--batch 2 \
--data data/original.yaml \
--img 640 \
--cfg models/detect/yolov9-t_original.yaml \
--weights best-t-sek.pt \
--name yolov9-shoulder-elbow-knee-t \
--hyp hyp.scratch-high_original.yaml \
--min-items 0 \
--epochs 100 \
--close-mosaic 15
- 700枚, T, 28クラス
Class Images Instances P R mAP50 mAP50-95
all 273 18058 0.602 0.462 0.495 0.338
body 273 1628 0.823 0.698 0.782 0.605
adult 273 1333 0.819 0.626 0.727 0.61
child 273 139 0.494 0.489 0.492 0.414
male 273 860 0.745 0.699 0.759 0.647
female 273 353 0.587 0.62 0.638 0.539
head 273 1394 0.846 0.829 0.871 0.606
front 273 246 0.492 0.512 0.504 0.397
right-front 273 322 0.573 0.404 0.443 0.353
right-side 273 181 0.626 0.459 0.506 0.398
right-back 273 72 0.425 0.431 0.427 0.335
back 273 87 0.319 0.299 0.212 0.159
left-back 273 73 0.321 0.31 0.305 0.222
left-side 273 169 0.632 0.426 0.443 0.341
left-front 273 244 0.452 0.299 0.321 0.249
face 273 776 0.851 0.785 0.834 0.556
eye 273 700 0.577 0.216 0.27 0.0923
nose 273 637 0.719 0.385 0.445 0.202
mouth 273 524 0.591 0.3 0.333 0.129
ear 273 590 0.68 0.378 0.425 0.216
hand 273 1046 0.847 0.597 0.699 0.424
hand_left 273 525 0.685 0.507 0.595 0.381
hand_right 273 521 0.679 0.491 0.565 0.343
foot 273 885 0.637 0.593 0.619 0.345
shoulder 273 2200 0.446 0.29 0.272 0.0943
elbow 273 1433 0.389 0.158 0.164 0.0565
knee 273 1120 0.387 0.213 0.212 0.0797
- 837枚, E, 3クラス
python train_dual.py \
--workers 8 \
--device 0 \
--batch 4 \
--data data/original.yaml \
--img 640 \
--cfg models/detect/yolov9-e_original.yaml \
--weights best-e.pt \
--name yolov9-shoulder-elbow-knee-e \
--hyp hyp.scratch-high_original.yaml \
--min-items 0 \
--epochs 200 \
--close-mosaic 15
Class Images Instances P R mAP50 mAP50-95
all 327 5679 0.82 0.579 0.674 0.382
shoulder 327 2524 0.83 0.597 0.696 0.386
elbow 327 1742 0.802 0.549 0.636 0.361
knee 327 1413 0.829 0.592 0.69 0.398
| image | image |
|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
PINTO| image | image |
|---|---|
 |
 |
 |
 |
PINTO- 1,000枚, E, 3クラス
python train_dual.py \
--workers 8 \
--device 0 \
--batch 4 \
--data data/original.yaml \
--img 640 \
--cfg models/detect/yolov9-e_original.yaml \
--weights best-e-sek.pt \
--name yolov9-shoulder-elbow-knee-e \
--hyp hyp.scratch-high_original.yaml \
--min-items 0 \
--epochs 100 \
--close-mosaic 15
python val_dual.py \
--data data/original.yaml \
--img 640 \
--batch 8 \
--conf 0.001 \
--iou 0.7 \
--device 0 \
--weights runs/train/yolov9-shoulder-elbow-knee-e/weights/last.pt \
--name yolov9_e_640_val
Class Images Instances P R mAP50 mAP50-95
all 391 7908 0.88 0.624 0.729 0.451
shoulder 391 3748 0.881 0.603 0.726 0.441
elbow 391 2330 0.866 0.602 0.697 0.434
knee 391 1830 0.892 0.665 0.764 0.479
| YOLOv9 | YOLOX + ViTPose |
|---|---|
 |
 |
| YOLOv9 | Sapiens |
|---|---|
|
 |
PINTO最終的に狙う姿。
