Ooredoo Stamp Detection Model

A fine-tuned YOLOS model for detecting stamps in documents.

Model Description

This model is based on hustvl/yolos-small and fine-tuned for single-class stamp detection in documents. It can identify and localize stamps with high accuracy.

Base Model: hustvl/yolos-small
Task: Object Detection
Classes: 1 (stamp)
Input: RGB images
Output: Bounding boxes + confidence scores

Author

V. Hisamutdinovs

Usage

from transformers import AutoImageProcessor, AutoModelForObjectDetection
from PIL import Image
import torch

# Load model
processor = AutoImageProcessor.from_pretrained("Ooredoo-Group/ooredoo-stamp-detection")
model = AutoModelForObjectDetection.from_pretrained("Ooredoo-Group/ooredoo-stamp-detection")

# Inference
image = Image.open("document_with_stamp.jpg")
inputs = processor(images=image, return_tensors="pt")
outputs = model(**inputs)

# Post-process results
target_sizes = torch.tensor([image.size[::-1]])
results = processor.post_process_object_detection(outputs, target_sizes=target_sizes)[0]

# Display results
for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
    print(f"Detected stamp with confidence {score:.2f}")
    print(f"Bounding box: {box}")

Training Data

Total Images: 350
Total Annotations: 711 bounding boxes
Format: COCO format
Split: Train (80%) / Val (15%) / Test (5%)
Sources: Kaggle stamp dataset + HuggingFace datasets

Performance

Confidence: High confidence detections (0.99+)
Accuracy: Excellent performance on document stamp detection
Speed: Fast inference suitable for production use

Limitations

Designed specifically for stamp detection
Performance may vary on different document types
Requires proper image preprocessing

Organization

Ooredoo-Group

License

Apache-2.0

Ooredoo-Group/ooredoo-stamp-detection