📸
最新のYOLOv8 が公開されたので試してみる
最新のYOLOv8がUltralyticsより公開されました。まだ論文がarxivに公開されていないらしく情報が少ないですが、とりあえず試してみることにします。
ドキュメントをパッと見てまず思ったのがYOLOv8がPythonのパッケージとして配布されていることですかね。https://docs.ultralytics.com によると,YOLOv8では以前のYOLOを切り替えることができるみたいです。これは拡張性とかかなり期待できそうです！
早速bus.jpgでテスト。
$ pip install ultralytics
$ yolo task=detect mode=predict model=yolov8n.pt source="https://ultralytics.com/images/bus.jpg"
認識されました。YOLOv5ではstop signの検出はされなかった記憶があるので性能アップしてるのかな？
detection
COCO val2017による事前学習済みモデルのmAPだけを比較すると検出の精度は軒並み上がってはいるようです。
YOLOv5
YOLOv8
yaml
https://roboflow.com/ のfootball-player-detection のデータセットを使って学習と推論を確かめてみました。付属されたyamlの内容なこのようになっています。クラス数は4で、ボール、ゴールキーパー、プレイヤー、レフリーとなっています。
data.yaml
names:
- ball
- goalkeeper
- player
- referee
nc: 4
roboflow:
license: CC BY 4.0
project: football-players-detection-3zvbc
url: https://universe.roboflow.com/roboflow-jvuqo/football-players-detection-3zvbc/dataset/4
version: 4
workspace: roboflow-jvuqo
test: ../test/images
train: football-players-detection-4/train/images
val: football-players-detection-4/valid/images
train
100エポックで学習させました。モデルは一番小さいやつです。
from ultralytics import YOLO
model = YOLO("yolov8n.pt")
model.train(data="./football-players-detection-4/data.yaml", batch=4) # train the model
学習のlogはこちら
yolo/engine/trainer: task=detect, mode=train, model=yolov8n.yaml, data=./football-players-detection-4/data.yaml, epochs=100, patience=50, batch=4, imgsz=640, save=True, cache=False, device=None, workers=4, project=None, name=None, exist_ok=False, pretrained=False, optimizer=SGD, verbose=False, seed=0, deterministic=True, single_cls=False, image_weights=False, rect=False, cos_lr=False, close_mosaic=10, resume=False, overlap_mask=True, mask_ratio=4, dropout=False, val=True, save_json=False, save_hybrid=False, conf=0.001, iou=0.7, max_det=300, half=True, dnn=False, plots=True, source=ultralytics/assets/, show=False, save_txt=False, save_conf=False, save_crop=False, hide_labels=False, hide_conf=False, vid_stride=1, line_thickness=3, visualize=False, augment=False, agnostic_nms=False, retina_masks=False, format=torchscript, keras=False, optimize=False, int8=False, dynamic=False, simplify=False, opset=17, workspace=4, nms=False, lr0=0.01, lrf=0.01, momentum=0.937, weight_decay=0.001, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=7.5, cls=0.5, dfl=1.5, fl_gamma=0.0, label_smoothing=0.0, nbs=64, hsv_h=0.015, hsv_s=0.7, hsv_v=0.4, degrees=0.0, translate=0.1, scale=0.5, shear=0.0, perspective=0.0, flipud=0.0, fliplr=0.5, mosaic=1.0, mixup=0.0, copy_paste=0.0, hydra={'output_subdir': None, 'run': {'dir': '.'}}, v5loader=True, save_dir=runs/detect/train13
Ultralytics YOLOv8.0.3 🚀 Python-3.10.6 torch-1.13.1+cu117 CUDA:0 (NVIDIA GeForce RTX 3060, 12288MiB)
Downloading https://ultralytics.com/assets/Arial.ttf to /root/.config/Ultralytics/Arial.ttf...
100%|█████████████████████████████████████████████████████████████████████| 755k/755k [00:01<00:00, 651kB/s]
Overriding model.yaml nc=80 with nc=4
from n params module arguments
0 -1 1 464 ultralytics.nn.modules.Conv [3, 16, 3, 2]
1 -1 1 4672 ultralytics.nn.modules.Conv [16, 32, 3, 2]
2 -1 1 7360 ultralytics.nn.modules.C2f [32, 32, 1, True]
3 -1 1 18560 ultralytics.nn.modules.Conv [32, 64, 3, 2]
4 -1 2 49664 ultralytics.nn.modules.C2f [64, 64, 2, True]
5 -1 1 73984 ultralytics.nn.modules.Conv [64, 128, 3, 2]
6 -1 2 197632 ultralytics.nn.modules.C2f [128, 128, 2, True]
7 -1 1 295424 ultralytics.nn.modules.Conv [128, 256, 3, 2]
8 -1 1 460288 ultralytics.nn.modules.C2f [256, 256, 1, True]
9 -1 1 164608 ultralytics.nn.modules.SPPF [256, 256, 5]
10 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
11 [-1, 6] 1 0 ultralytics.nn.modules.Concat [1]
12 -1 1 148224 ultralytics.nn.modules.C2f [384, 128, 1]
13 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest']
14 [-1, 4] 1 0 ultralytics.nn.modules.Concat [1]
15 -1 1 37248 ultralytics.nn.modules.C2f [192, 64, 1]
16 -1 1 36992 ultralytics.nn.modules.Conv [64, 64, 3, 2]
17 [-1, 12] 1 0 ultralytics.nn.modules.Concat [1]
18 -1 1 123648 ultralytics.nn.modules.C2f [192, 128, 1]
19 -1 1 147712 ultralytics.nn.modules.Conv [128, 128, 3, 2]
20 [-1, 9] 1 0 ultralytics.nn.modules.Concat [1]
21 -1 1 493056 ultralytics.nn.modules.C2f [384, 256, 1]
22 [15, 18, 21] 1 752092 ultralytics.nn.modules.Detect [4, [64, 128, 256]]
Model summary: 225 layers, 3011628 parameters, 3011612 gradients, 8.2 GFLOPs
Transferred 319/355 items from pretrained weights
optimizer: SGD(lr=0.01) with parameter groups 57 weight(decay=0.0), 64 weight(decay=0.001), 63 bias
train: Scanning /work/football-players-detection-4/train/labels.cache... 204 images, 0 backgrounds, 0 corrup
val: Scanning /work/football-players-detection-4/valid/labels.cache... 38 images, 0 backgrounds, 0 corrupt:
Image sizes 640 train, 640 val
Using 4 dataloader workers
Logging results to runs/detect/train13
Starting training for 100 epochs...
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
1/100 1.47G 2.009 3.621 0.9438 165 640: 100%|██████████| 51/51 [00:11
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.0338 0.144 0.121 0.0373
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
2/100 1.73G 2.336 2.039 0.9439 162 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.88 0.127 0.108 0.034
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
3/100 1.73G 2.742 2.028 1.004 178 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.877 0.109 0.106 0.0309
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
4/100 1.73G 2.605 1.859 1.019 192 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.9 0.176 0.166 0.055
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
5/100 1.73G 2.611 1.756 0.9955 205 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.893 0.144 0.143 0.047
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
6/100 1.99G 2.384 1.643 0.9691 134 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.644 0.181 0.154 0.0492
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
7/100 1.99G 2.376 1.568 0.9644 165 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.679 0.196 0.182 0.0577
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
8/100 1.99G 2.3 1.529 0.953 259 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.915 0.176 0.175 0.0616
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
9/100 1.99G 2.12 1.494 0.9399 205 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.674 0.209 0.18 0.0781
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
10/100 1.99G 2.163 1.433 0.9156 140 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.888 0.189 0.158 0.0566
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
11/100 1.99G 2.047 1.386 0.911 206 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.945 0.198 0.217 0.0912
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
12/100 1.99G 1.966 1.318 0.9019 201 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.813 0.224 0.237 0.111
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
13/100 1.99G 1.883 1.272 0.8939 120 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.76 0.254 0.259 0.128
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
14/100 1.99G 1.866 1.281 0.8882 83 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.742 0.256 0.242 0.106
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
15/100 1.99G 1.85 1.255 0.897 116 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.892 0.26 0.291 0.134
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
16/100 1.99G 1.84 1.185 0.8925 171 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.885 0.265 0.303 0.155
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
17/100 1.99G 1.855 1.187 0.8838 168 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.862 0.262 0.309 0.158
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
18/100 1.99G 1.823 1.15 0.8752 193 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.837 0.263 0.307 0.156
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
19/100 1.99G 1.674 1.105 0.8691 175 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.828 0.263 0.311 0.169
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
20/100 1.99G 1.749 1.106 0.8696 107 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.84 0.281 0.333 0.164
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
21/100 1.99G 1.716 1.105 0.8694 205 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.519 0.315 0.345 0.181
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
22/100 1.99G 1.702 1.087 0.8648 108 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.889 0.29 0.357 0.175
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
23/100 1.99G 1.655 1.052 0.8674 195 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.918 0.287 0.363 0.195
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
24/100 1.99G 1.67 1.074 0.8647 209 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.908 0.28 0.375 0.205
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
25/100 1.99G 1.628 1.013 0.8636 211 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.911 0.285 0.37 0.202
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
26/100 1.99G 1.618 1.004 0.8611 141 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.813 0.33 0.368 0.187
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
27/100 1.99G 1.659 1.001 0.8645 120 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.832 0.323 0.374 0.198
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
28/100 1.99G 1.59 0.9911 0.8556 176 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.802 0.345 0.397 0.213
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
29/100 1.99G 1.524 0.9619 0.8515 131 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.663 0.41 0.399 0.21
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
30/100 1.99G 1.562 0.9773 0.8536 303 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.676 0.409 0.407 0.219
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
31/100 1.99G 1.505 0.9305 0.8442 178 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.652 0.403 0.423 0.223
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
32/100 1.99G 1.568 0.9467 0.8507 78 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.681 0.421 0.417 0.229
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
33/100 1.99G 1.478 0.9109 0.8463 192 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.72 0.442 0.456 0.26
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
34/100 1.99G 1.477 0.8861 0.8472 160 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.734 0.451 0.477 0.251
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
35/100 1.99G 1.46 0.8981 0.8489 239 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.734 0.444 0.508 0.288
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
36/100 1.99G 1.478 0.8897 0.8439 218 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.782 0.462 0.527 0.305
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
37/100 1.99G 1.397 0.8607 0.8461 135 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.785 0.513 0.539 0.301
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
38/100 1.99G 1.45 0.8713 0.8445 210 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.799 0.519 0.556 0.341
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
39/100 1.99G 1.484 0.8736 0.8436 238 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.772 0.496 0.559 0.302
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
40/100 1.99G 1.434 0.8443 0.8446 176 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.787 0.534 0.568 0.336
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
41/100 1.99G 1.365 0.8316 0.8407 147 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.833 0.501 0.561 0.323
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
42/100 1.99G 1.399 0.8436 0.8373 209 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.825 0.517 0.565 0.31
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
43/100 1.99G 1.381 0.8311 0.8404 159 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.849 0.502 0.572 0.316
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
44/100 1.99G 1.359 0.8052 0.8392 115 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.841 0.527 0.586 0.356
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
45/100 1.99G 1.409 0.8311 0.8351 230 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.897 0.507 0.593 0.335
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
46/100 1.99G 1.39 0.8033 0.8406 127 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.881 0.521 0.596 0.352
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
47/100 1.99G 1.397 0.8045 0.8387 147 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.792 0.559 0.587 0.342
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
48/100 1.99G 1.308 0.7802 0.8309 97 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.853 0.555 0.609 0.367
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
49/100 1.99G 1.351 0.7775 0.8326 130 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.884 0.556 0.611 0.345
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
50/100 1.99G 1.371 0.7948 0.8323 135 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.909 0.557 0.62 0.371
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
51/100 1.99G 1.324 0.7599 0.8291 114 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.888 0.575 0.617 0.362
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
52/100 1.99G 1.333 0.7634 0.8313 176 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.859 0.576 0.614 0.357
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
53/100 1.99G 1.316 0.7678 0.8298 89 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.864 0.558 0.616 0.377
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
54/100 1.99G 1.304 0.7515 0.829 205 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.836 0.561 0.612 0.365
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
55/100 1.99G 1.291 0.7561 0.8314 199 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.849 0.588 0.617 0.385
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
56/100 1.99G 1.28 0.745 0.8277 145 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.796 0.598 0.609 0.381
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
57/100 1.99G 1.265 0.7361 0.8292 159 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.816 0.59 0.616 0.391
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
58/100 1.99G 1.316 0.7579 0.8236 184 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.856 0.568 0.614 0.379
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
59/100 1.99G 1.307 0.7426 0.8248 201 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.829 0.589 0.613 0.373
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
60/100 1.99G 1.263 0.7287 0.8249 196 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.804 0.613 0.611 0.377
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
61/100 1.99G 1.261 0.7306 0.8294 153 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.836 0.573 0.617 0.385
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
62/100 1.99G 1.242 0.7202 0.8309 141 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.848 0.583 0.614 0.379
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
63/100 1.99G 1.229 0.7055 0.8281 247 640: 100%|██████████| 51/51 [00:07
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.854 0.542 0.606 0.366
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
64/100 1.99G 1.226 0.7115 0.8258 212 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.853 0.597 0.622 0.399
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
65/100 1.99G 1.219 0.7067 0.8207 132 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.842 0.593 0.619 0.395
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
66/100 1.99G 1.229 0.7119 0.8277 118 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.843 0.58 0.625 0.392
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
67/100 1.99G 1.222 0.7114 0.827 131 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.845 0.589 0.629 0.398
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
68/100 1.99G 1.237 0.7146 0.8227 79 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.84 0.564 0.617 0.388
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
69/100 1.99G 1.221 0.6929 0.8193 118 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.873 0.566 0.635 0.409
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
70/100 1.99G 1.178 0.6823 0.8187 155 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.892 0.586 0.634 0.396
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
71/100 1.99G 1.221 0.7006 0.8204 178 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.863 0.594 0.627 0.402
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
72/100 1.99G 1.179 0.6768 0.8146 170 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.866 0.57 0.629 0.395
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
73/100 1.99G 1.193 0.6817 0.8214 109 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.861 0.585 0.628 0.404
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
74/100 1.99G 1.167 0.6858 0.8179 168 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.861 0.597 0.63 0.399
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
75/100 1.99G 1.186 0.6846 0.8165 157 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.872 0.599 0.635 0.398
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
76/100 1.99G 1.167 0.6659 0.8174 142 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.866 0.614 0.633 0.4
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
77/100 1.99G 1.154 0.6747 0.8187 171 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.856 0.612 0.629 0.391
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
78/100 1.99G 1.117 0.6599 0.8153 90 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.862 0.605 0.628 0.397
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
79/100 1.99G 1.174 0.677 0.8133 145 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.854 0.606 0.633 0.407
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
80/100 1.99G 1.161 0.6688 0.8162 201 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.841 0.611 0.631 0.407
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
81/100 1.99G 1.142 0.658 0.8184 201 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.833 0.617 0.631 0.407
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
82/100 1.99G 1.143 0.6509 0.8116 150 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.84 0.628 0.626 0.405
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
83/100 1.99G 1.138 0.6508 0.8134 185 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.855 0.632 0.636 0.414
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
84/100 1.99G 1.159 0.6615 0.8112 249 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.86 0.618 0.647 0.42
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
85/100 1.99G 1.125 0.6503 0.8129 152 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.864 0.611 0.646 0.417
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
86/100 1.99G 1.112 0.6412 0.8154 170 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.856 0.609 0.639 0.415
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
87/100 1.99G 1.149 0.647 0.8126 205 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.855 0.614 0.635 0.415
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
88/100 1.99G 1.093 0.6269 0.811 177 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.856 0.609 0.636 0.416
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
89/100 1.99G 1.087 0.6333 0.8119 169 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.848 0.62 0.638 0.413
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
90/100 1.99G 1.12 0.6398 0.8129 169 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.88 0.591 0.649 0.42
Closing dataloader mosaic
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
91/100 1.99G 1.064 0.6786 0.8115 90 640: 100%|██████████| 51/51 [00:06
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.844 0.591 0.647 0.407
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
92/100 1.99G 1.035 0.6426 0.8113 90 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.854 0.583 0.642 0.411
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
93/100 1.99G 1.061 0.6618 0.805 95 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.849 0.589 0.639 0.409
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
94/100 1.99G 1.062 0.6508 0.8099 91 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.843 0.606 0.646 0.415
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
95/100 1.99G 1.05 0.6432 0.8106 91 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.835 0.616 0.654 0.417
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
96/100 1.99G 1.059 0.6453 0.8104 93 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.848 0.622 0.655 0.421
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
97/100 1.99G 1.033 0.6385 0.8024 91 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.833 0.633 0.65 0.421
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
98/100 1.99G 1.017 0.6259 0.8074 98 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.83 0.627 0.651 0.426
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
99/100 1.99G 1.015 0.623 0.8082 89 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.834 0.623 0.651 0.428
Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
100/100 1.99G 1.031 0.6333 0.8088 92 640: 100%|██████████| 51/51 [00:05
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.839 0.626 0.65 0.422
100 epochs completed in 0.260 hours.
Optimizer stripped from runs/detect/train13/weights/last.pt, 6.3MB
Optimizer stripped from runs/detect/train13/weights/best.pt, 6.3MB
Validating runs/detect/train13/weights/best.pt...
Ultralytics YOLOv8.0.3 🚀 Python-3.10.6 torch-1.13.1+cu117 CUDA:0 (NVIDIA GeForce RTX 3060, 12288MiB)
Fusing layers...
Model summary: 168 layers, 3006428 parameters, 0 gradients, 8.1 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 60%|██████ | 3WARNING ⚠️ NMS time limit 0.900s exceeded
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 5
all 38 905 0.832 0.596 0.624 0.407
ball 38 35 1 0 0 0
goalkeeper 38 27 0.812 0.802 0.877 0.574
player 38 754 0.871 0.895 0.922 0.634
referee 38 89 0.646 0.685 0.696 0.421
Speed: 1.6ms pre-process, 56.7ms inference, 0.0ms loss, 53.7ms post-process per image
Saving runs/detect/train13/predictions.json...
Results saved to runs/detect/train13
テスト用の画像
テスト前の画像を一枚だけ載せます。ドルトムントの試合みたいですね！
検出結果
player だけでなく referee, goalkeeper も検出してくれてました。良い感じではないでしょうか。
!yolo task=detect mode=predict model=best.pt source="./football-players-detection-4/test/images"
ただ、別の画像の検出ではボールまでは捉えてくれませんでした。選手よりだいぶ小さいので難しいのですかね？
感想
面白かったです。次はサッカーのリアルタイム動画像でどう検出されるのか試すことにします。
Discussion
記事のとうりにやってみて、できました。
ありがとうございます。
データをダウンロードするところのインストラクションもあると初心者にはやさしいです。