WARAJA/Tzefa-Line-Segmentation-YOLO

Tzefa Line Segmentation Model (YOLO11x-OBB)

Custom-trained oriented bounding box (OBB) model for detecting text lines in handwritten code images.

Architecture

• Model: YOLO11x-OBB (Ultralytics)
• Task: Oriented Bounding Box detection
• Classes: 1 (text_line)
• Input: RGB image (letterboxed to 640x640 internally)
• Output: OBB coordinates per text line, sorted top-to-bottom

Usage

from huggingface_hub import hf_hub_download
from ultralytics import YOLO

weights = hf_hub_download("WARAJA/Tzefa-Line-Segmentation-YOLO", "best.pt")
model = YOLO(weights)
results = model.predict("your_image.png", imgsz=640, conf=0.2, iou=0.2)

Training

• See training_artifacts/ for training curves, confusion matrices, and configuration.

Related

• Full Tzefa Pipeline
• OCR Demo
• Line Segmentation Dataset