YOLO11n det cvat2 project 1 重訓 取代 person_seg_yolo11n 訓練日期:2026-04-29 | 來源:cvat2 project 1(572 tasks,30K imgs / 154K bbox)
cvat2 project 1(person_v20260423):572 個 tasks,跨 GUANXI / CPC_OIL / RAICVAT P2 P26 / forklift_persons 等多種來源(包含 user 已 curate 過的高品質標注)。 本次訓練只取 bbox(polygon 自動轉 bbox)。
| Split | Imgs | BBoxes |
|---|---|---|
| Train | 22,131 | 114,844 |
| Validation | 3,508 | 19,061 |
| Test | 4,396 | 20,203 |
| 合計 | 30,035 | 154,108 |
from ultralytics import YOLO
model = YOLO("person_yolo11n_v20260502.pt")
results = model("image.jpg", conf=0.35)
for r in results:
for box in r.boxes:
x1, y1, x2, y2 = box.xyxy[0].tolist()
print(f"person {float(box.conf[0]):.2f} ({x1:.0f},{y1:.0f},{x2:.0f},{y2:.0f})")
本模型取代之前 model_viewer 中三個 handler 共用的 person YOLO:
已部署 http://192.168.53.21:7860/。 選任一 person-attr 模型(ppe21 / safety_rope / forklift_ppe)就會用本模型偵測 person。
yolo detect train \
data=/mnt/ssd/cvat2/external/person_yolo_v20260502/data.yaml \
model=yolo11n.pt \
epochs=80 imgsz=640 batch=32 device=1 \
project=/home/ubuntu/person_runs name=v20260502 \
patience=15
| Backbone | YOLO11n(pretrained on COCO) |
|---|---|
| Optimizer | SGD(ultralytics 預設) |
| Batch | 32 |
| Image size | 640 × 640 |
| Epochs run | 80 / 80(patience=15 未觸發) |
| Best epoch | 79(mAP50-95=0.468) |
| 訓練時間 | ~87 分鐘(1× RTX 5090) |
| AMP | auto(fp16) |
| Augmentation | HSV、flip、mosaic、close-mosaic last 10 epoch(ultralytics 預設) |
R2 Bucket:rai-models / person_yolo11n_v20260502 / best.pt(待上傳)
本機:/Users/davidyang/code/2026/模型訓練/scripts/model_viewer/models/person_yolo11n_v20260502.pt
Generated 2026-04-29 | 回到目錄