Yolov8 Nano Aadhar Card

🚀 YOLOv8 model to detect import texts on an Aadhar Card

This project uses the YOLOv8 model to detect important texts on an Aadhaar Card, which can be applied in scenarios like automatic data entry, fraud detection, and document verification.

🚀 Quick Start

Aadhaar Card text detection is the process of identifying and extracting text from Aadhaar Card images. This can be useful for a variety of applications, such as automatic data entry, fraud detection, and document verification.

One approach to Aadhaar Card text detection is to use YOLOv8, a state-of-the-art object detection model. YOLOv8 can be trained to detect a variety of object classes, including text. Once trained, YOLOv8 can be used to detect text in Aadhaar Card images and extract the text to a text file or other format.

✨ Features

• This model can detect important texts on an Aadhaar Card, including AADHAR_NUMBER, DATE_OF_BIRTH, GENDER, NAME, and ADDRESS.
• It provides high precision and recall in text detection, with mAP@50 reaching 0.963 and mAP@50 - 95 reaching 0.748.

📦 Installation

Install Dependencies

$ pip install ultralytics huggingface_hub supervision

💻 Usage Examples

Basic Usage

# label_id: label_name
{0: "AADHAR_NUMBER", 1: "DATE_OF_BIRTH", 2: "GENDER", 3: "NAME", 4: "ADDRESS"}

Advanced Usage

from ultralytics import YOLO
from huggingface_hub import hf_hub_download
from supervision import Detections

# repo details
repo_config = dict(
    repo_id = "arnabdhar/YOLOv8-nano-aadhar-card",
    filename = "model.pt",
    local_dir = "./models"
)

# load model
model = YOLO(hf_hub_download(**repo_config))

# get id to label mapping
id2label = model.names
print(id2label)

# Perform Inference
image_url = "https://i.pinimg.com/originals/08/6d/82/086d820550f34066764f4047ddc263ca.jpg"

detections = Detections.from_ultralytics(model.predict(image_url)[0])

print(detections)

🔧 Technical Details

Fine Tuning

The following hyperparameters were used to finetune the model

model: yolov8n.pt
batch: 4
epochs: 100
optimizer: AdamW
warmup_epochs: 15
seed: 42
imgsz: 640

The following evaluation metrics were achieved by best.pt for bounding box predictions:

recall: 0.962
precision: 0.973
mAP50: 0.963
mAP50_95: 0.748 

Dataset

📄 License

This project is licensed under the Apache-2.0 license.

📚 Documentation

Model Index