Roboflow Supervision 完整教學：用 Python 輕鬆做物件追蹤、區域計數與影片分析（2026）

pip install supervision

pip install supervision ultralytics opencv-python

import supervision as sv
from ultralytics import YOLO

model = YOLO("yolov8n.pt")
result = model("path/to/image.jpg")[0]

# 關鍵：轉換成 supervision 統一格式
detections = sv.Detections.from_ultralytics(result)

print(len(detections))       # 偵測到幾個物件
print(detections.xyxy)       # 邊框座標
print(detections.class_id)   # 類別 ID

import supervision as sv
from PIL import Image
from rfdetr import RFDETRSmall

image = Image.open("path/to/image.jpg")
model = RFDETRSmall()
detections = model.predict(image, threshold=0.5)

print(len(detections))  # 例如 5

import cv2
import supervision as sv

image = cv2.imread("path/to/image.jpg")
# 假設 detections 來自模型
detections = sv.Detections(...)

box_annotator = sv.BoxAnnotator()
label_annotator = sv.LabelAnnotator()

annotated = box_annotator.annotate(scene=image.copy(), detections=detections)
annotated = label_annotator.annotate(scene=annotated, detections=detections)

import supervision as sv
from ultralytics import YOLO

model = YOLO("yolov8n.pt")
tracker = sv.ByteTrack()          # 初始化 ByteTrack

video_info = sv.VideoInfo.from_video_path("input.mp4")
trace_annotator = sv.TraceAnnotator()
box_annotator = sv.BoxAnnotator()

with sv.VideoSink("output.mp4", video_info) as sink:
    for frame in sv.get_video_frames_generator("input.mp4"):
        result = model(frame)[0]
        detections = sv.Detections.from_ultralytics(result)

        # 關鍵：把偵測結果交給 ByteTrack，取得追蹤 ID
        detections = tracker.update_with_detections(detections)

        annotated = trace_annotator.annotate(scene=frame, detections=detections)
        annotated = box_annotator.annotate(scene=annotated, detections=detections)

        sink.write_frame(annotated)

import supervision as sv
from ultralytics import YOLO
import numpy as np

model = YOLO("yolov8n.pt")
tracker = sv.ByteTrack()

# 定義感興趣區域（多邊形頂點座標，順時針或逆時針皆可）
zone_polygon = np.array([
    [100, 400], [600, 400], [600, 600], [100, 600]
])
zone = sv.PolygonZone(polygon=zone_polygon)
zone_annotator = sv.PolygonZoneAnnotator(
    zone=zone, color=sv.Color.GREEN
)

for frame in sv.get_video_frames_generator("input.mp4"):
    result = model(frame)[0]
    detections = sv.Detections.from_ultralytics(result)
    detections = tracker.update_with_detections(detections)

    # 關鍵：過濾出「區域內」的物件
    zone.trigger(detections)
    print(f"區域內物件數：{len(zone.current_count)}")

    annotated = zone_annotator.annotate(scene=frame)
    # ...輸出到 VideoSink

import supervision as sv

line_start = sv.Point(50, 400)   # 線段起點
line_end = sv.Point(1250, 400)   # 線段終點

line = sv.LineZone(start=line_start, end=line_end)
line_annotator = sv.LineZoneAnnotator(
    thickness=4, text_thickness=4, text_scale=2
)

# 在每一幀中：line.trigger(detections) 即可自動更新計數
# line.in_count / line.out_count 分別是「進」與「出」的次數

import supervision as sv

# 讀取 COCO 格式
ds = sv.DetectionDataset.from_coco(
    images_directory_path="data/train",
    annotations_path="data/train/_annotations.coco.json",
)

# 分割成 train / test / valid
train, rest = ds.split(split_ratio=0.7)
test, valid = rest.split(split_ratio=0.5)

# 轉換成 YOLO 格式匯出
ds.as_yolo(
    images_directory_path="yolo_out/images",
    annotations_directory_path="yolo_out/labels",
    data_yaml_path="yolo_out/data.yaml",
)