有没有PaddleX 3.0以上版本的目标跟踪产线或模块的示例

[wyd1520]

有没有PaddleX 3.0以上版本的目标跟踪产线或模块的示例，找了好久没找到，产线的配置文件也没有。只看到在PaddleDetection2.0上有

[liuhongen1234567]

您好，paddlex3.0目前暂不支持目标跟踪产线。

[wyd1520]

您好，paddlex3.0目前暂不支持目标跟踪产线。

谢谢回复，不过我用了这样的方式实现的调用 ppdet下的JDETracker，现在还有一个ReID的跨镜头跟踪没有实现，缺失相关的文档说明。主是就是update的第二个参数pred_embs，ReID特征是用什么模型提取的。官网没有找到这个说明文档

import time

import cv2
import numpy as np
from ppdet.modeling import JDETracker

from paddlex import create_pipeline, create_model

model = create_model(model_name="PP-YOLOE-L_vehicle")
# model = create_predictor(model_name="PicoDet-S")
print(model.config.get("label_list"))
# 2. 创建 JDETracker
tracker = JDETracker(use_byte=True, num_classes=len(model.config.get("label_list")))

cap = cv2.VideoCapture("../../test_video/vehicles.mp4")
while cap.isOpened():
    success, frame = cap.read()
    frame = cv2.resize(frame, dsize=None, fx=0.5, fy=0.5)

    # 4. PaddleX 检测
    results = model.predict(frame, batch_size=1, threshold=0.5)
    for res in results:
        boxes = res['boxes']  # list of dict: {"coordinate": [x0,y0,x1,y1], "score": float, "label": str, "cls_id": int}
        if len(boxes) == 0:
            pred_dets = np.zeros((0, 6), dtype=np.float32)
        else:
            pred_dets = []
            for b in boxes:
                cls_id = b["cls_id"]
                score = b["score"]
                x0, y0, x1, y1 = b["coordinate"]
                pred_dets.append([cls_id, score, x0, y0, x1, y1])
            pred_dets = np.array(pred_dets, dtype=np.float32)

        # 5. 更新 tracker
        tracks = tracker.update(pred_dets)

        # 6. 可视化
        for cls_id, track_list in tracks.items():
            for t in track_list:
                tlwh = t.tlwh  # left, top, w, h
                x, y, w, h = map(int, tlwh)
                track_id = t.track_id
                cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0), 2)
                cv2.putText(frame, f"ID:{track_id}", (x, y - 5),
                            cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1)

    cv2.imshow("tracking", frame)
    if cv2.waitKey(1) & 0xFF == 27:
        break

cap.release()
cv2.destroyAllWindows()