Views
No views yet
telecom_videos set (fixed overhead/elevated cameras).| Architecture | YOLOv11s (Ultralytics) |
| Input | 640×640 RGB |
| Output | output0 shape [1, 14, N] — 4 bbox + 10 class scores |
| Parameters | 9.5 M |
| Weights | best.pt (19 MB) / best.onnx (37 MB, opset 19) |
| Epochs trained | 67 (early-stop from 80) |
| Best epoch | 52 |
| Val mAP50 | 0.377 |
| Val mAP50-95 | 0.219 |
| Val Precision | 0.502 |
| Val Recall | 0.388 |
pedestrian, people, bicycle, car, van, truck, tricycle,
awning-tricycle, bus, motor — matching the VisDrone2019 taxonomy.data=VisDrone.yaml (Ultralytics).1yolo detect train data=VisDrone.yaml model=yolo11s.pt \
2 epochs=80 imgsz=640 batch=32 patience=15 device=01from ultralytics import YOLO
2model = YOLO("best.pt")
3r = model("traffic_camera_frame.jpg", conf=0.25)
4r[0].show()1import cv2, numpy as np, onnxruntime as ort
2
3sess = ort.InferenceSession("best.onnx", providers=["CUDAExecutionProvider"])
4img = cv2.imread("frame.jpg")
5# letterbox to 640×640, RGB, [0,1], CHW
6scale = min(640/img.shape[1], 640/img.shape[0])
7nw, nh = int(img.shape[1]*scale), int(img.shape[0]*scale)
8r = cv2.resize(img, (nw, nh))
9canvas = np.full((640, 640, 3), 114, np.uint8)
10pad_x, pad_y = (640-nw)//2, (640-nh)//2
11canvas[pad_y:pad_y+nh, pad_x:pad_x+nw] = r
12x = cv2.cvtColor(canvas, cv2.COLOR_BGR2RGB).astype(np.float32)/255.0
13x = np.ascontiguousarray(x.transpose(2,0,1)[None])
14y = sess.run(None, {"images": x})[0][0] # (14, N)
15# decode: y[:4] = cx,cy,w,h; y[4:14] = class scorescar).