Tracker giving unexpected error on colab

[tanmy-sharma]

I installed boxmot on colab using pip install boxmot. I was using L4 GPU on colab pro. Please let me know how to fix this issue.

[mikel-brostrom]

I updated the argument names in README. I made some changes that were not reflected there

[RaffelRavionaldo]

The code looks like this :

while True:
        ret, frame = vid.read()
        if not ret:
            break

        results = yolo_model.predict(frame, verbose=False, device=device)
        dets = []
        
        for r in results:
            boxes = r.boxes
            for box in boxes:
                xyxy = box.xyxy.cpu().numpy().flatten()  
                conf = float(box.conf.cpu().numpy()) 
                label = int(box.cls.cpu().numpy())
                dets.append([*xyxy, conf, label])

        dets = np.array(dets)

        if len(boxes) == 0:
            print('True')
            dets = np.empty((0,5))
            
        res = tracker.update(dets, frame)
        
        tracker.plot_results(frame, show_trajectories=True)

actually, I already tried to use your code, but still have the same error. May I see your parameters? maybe it's the problems (already tuned max_age, Q_xy_scaling, etc but still error) For me looks like this :

{
    "per_class": true,
    "fp16": false,
    "device": "cuda:0",
    "det_thresh": 0.25,
    "aw_param": 0.7,
    "embedding_off": false,
    "cmc_off": true,
    "aw_off": false,
    "half": false,
    "delta_t": 3,
    "iou_threshold": 0.7,
    "alpha_fixed_emb": 0.6,
    "min_hits": 1,
    "association_func": "iou",
    "inertia": 0.3,
    "Q_xy_scaling": 0.001,
    "Q_s_scaling": 1e-05,
    "reid_weights": "reid_weights/resnet50_fc512_msmt_xent.pth",
    "max_age": 80
}

[tanmy-sharma]

yolo model from ultralytics?

[tanmy-sharma]

I just tried and am unable to replicate the error. Please try the following code:

import numpy as np
import cv2
from boxmot import DeepOcSort
from pathlib import Path

tracker = DeepOcSort(
    reid_weights = Path('lmbn_n_cuhk03_d.pt'),
    device = 'cpu',
    half = False
)

cap = cv2.VideoCapture('/content/video.mp4')

frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
fps = cap.get(cv2.CAP_PROP_FPS)

output_file = 'output.mp4'
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out = cv2.VideoWriter(output_file, fourcc, fps, (frame_width, frame_height))

while True:
    ret, im = cap.read()
    if not ret:
        print("Reached the end of the video.")
        break

    results = model.predict(im, verbose = False)[0]
    boxes = results.boxes.xyxy.cpu().numpy()  # Convert to numpy array (x_min, y_min, x_max, y_max)
    confs = results.boxes.conf.cpu().numpy()  # Get confidence scores
    classes = results.boxes.cls.cpu().numpy()

    if len(boxes) > 0:
        dets = np.hstack([boxes, confs[:, np.newaxis], classes[:, np.newaxis]])
        tracker.update(dets, im)
    else:
        dets = np.empty((0, 6))
        tracker.update(dets, im)

    tracker.plot_results(im, show_trajectories=True)
    out.write(im)

cap.release()
out.release()
cv2.destroyAllWindows()

[RaffelRavionaldo]

Yes, I am using yolo from Ultralytics. and Thanks for the code. It worked for me.