I need to track an object in a video, but I'm having some problems.

[neiloe8023]

Search before asking

• I have searched the Supervision issues and found no similar feature requests.

Question

The first two frames both correctly identify the object and successfully draw the track.

However, starting from the second frame, detections remains at 0, and I'm unsure why.

I then tried using the tracker.reset() function, which worked, but the track lines were very messy.

Here's my code:

model = YOLO('BTrain09301516.pt')

tracker = sv.ByteTrack(frame_rate=25)
box_annotator = sv.BoxAnnotator(color_lookup=sv.ColorLookup.TRACK)
label_annotator = sv.LabelAnnotator(color_lookup=sv.ColorLookup.TRACK)
trace_annotator = sv.TraceAnnotator(trace_length=30, color_lookup=sv.ColorLookup.TRACK)


image_dir = Path("images")
img_suffixes = ('*.jpg', '*.jpeg')
img_paths = []
for suf in img_suffixes:
    img_paths.extend(image_dir.glob(suf))
img_paths = sorted(img_paths)

for idx, img_path in enumerate(img_paths, 1):
    frame = cv2.imread(str(img_path))
    results = model(frame, verbose=False, conf=0.25)[0]
    detections = sv.Detections.from_ultralytics(results) # 转成 supervision 格式

    detections = tracker.update_with_detections(detections)
    print(f"[{idx}/{len(img_paths)}] ball>> results: {len(results)}, detections: {len(detections)}, conf: {results.boxes.conf.cpu().numpy()}")

    annotated_frame = frame.copy()
    annotated_frame = box_annotator.annotate(annotated_frame, detections)
    annotated_frame = label_annotator.annotate(annotated_frame, detections)
    annotated_frame = trace_annotator.annotate(annotated_frame, detections)

    cv2.imshow('frame', annotated_frame)
    tracker.reset() # test

    if cv2.waitKey(0) & 0xFF == ord('q'):
        cv2.destroyAllWindows()

---

The directory contains 300 images, which I extracted from video frames. I tried using the video, but the problem persisted starting with the second frame. I also tried modifying the ByteTrack parameters, but to no avail.

Please help me. I've been working on this for two days. I can display the example video from the official website: https://media.roboflow.com/supervision/video-examples/people-walking.mp4

Yolo can correctly infer objects.

Additional

No response

[neiloe8023]

I saw the issue mentioned in #1215 and tried using detections.xyxy = sv.pad_boxes(detections.xyxy, px=30, py=30) . It worked, but there were still instances of target loss (the confidence score of pt was very high, 0.80+). This might be due to insufficient frame rate.