YOLOv12_finetune_PortVehicleOCR

Model

YOLOv12 was fine-tuned on the maianh511/PortVehicleOCR dataset for object detection, specifically to localize license_plate (and container) regions in port vehicle images.

• Base model: YOLOv12
• Dataset: maianh511/PortVehicleOCR — 18,485 port vehicle images, 19,754 bounding boxes across 2 classes (container, license_plate), normalized to 640×640 px.
• Task: Object detection — locating the bounding box of license plates (and containers) in an image, as the localization step prior to OCR/text recognition.

Objective

Detect and localize license_plate regions accurately in real-world port images, providing cropped regions that feed into a downstream OCR pipeline (e.g., PP-OCRv4) for text recognition.

Usage

from ultralytics import YOLO

model = YOLO("path/to/finetuned_yolov12.pt")
results = model.predict(source="your_image.jpg", conf=0.25)

for r in results:
    r.show()      # visualize detections
    r.save()      # save annotated image
    print(r.boxes)  # bounding box coordinates + class + confidence