bakhil-aissa/YOLO-11-invoice_parsing

YOLO Model Card

Model Details

• Model Name: YOLO
• Version: 11
• Framework: Ultralytics YOLO /
• License: MIT
• Model Type: Object Detection

---

Intended Use

• Primary use case: Invoice parsing – detecting fields such as dates, totals, company names,
• Out of scope: Detecting non-invoice documents, handwriting recognition

---

Model Architecture

• Base Architecture: YOLOv11
• Input Resolution: 640 × 640

---

Dataset

• Source: https://universe.roboflow.com/davis-stepanskis/invoice-lbv9f/dataset/2

• Size: 6671 images

• Classes:

  {0: 'Discount_Percentage', 1: 'Due_Date', 2: 'Email_Client', 3: 'Name_Client', 4: 'Products', 5: 'Remise', 6: 'Subtotal', 7: 'Tax', 8: 'Tax_Precentage', 9: 'Tel_Client', 10: 'billing address', 11: 'header', 12: 'invoice date', 13:

'invoice number', 14: 'shipping address', 15: 'total'}

• Preprocessing:
  • Image resizing: 640 × 640 with letterboxing

---

Performance

• Evaluation Metrics:

  • mAP@[.5:.95]: 0.624.
  • Precision: 0.9160
  • Recall: 0.903

• Hardware Used:T4 x 2

---

Limitations

• Sensitive to document layout variations
• May fail on low-resolution or handwritten invoices
• Not robust to domain shifts (different invoice templates not seen during training)
• Requires OCR post-processing for text extraction

---

Ethical Considerations

• Biases: May not generalize well to invoices from different countries/languages
• Privacy: Invoices often contain personal and financial information – ensure compliance with GDPR/CCPA before using this model

---

How to Use

from ultralytics import YOLO

# Load model
model = YOLO("best.pt")

# Run inference
results = model.predict("invoice.jpg")

# Access results
for box in results[0].boxes:
    print(box.xyxy, box.conf, box.cls)