Sign recognition using YOLOv8, ABCNetv2, VietOCR, LLMs, ELECTRA-base Vietnamese

Guide to run Project

After pull project from Github, step-by-step to run project

1. Download checkpoints

Download all model and move to direction OCR/weights/ from link

2. Config variable

1. Access and get API KEY CLERK at link
2. Paste API Key to /web/.env.local

NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=...
CLERK_SECRET_KEY=...

3. Create file /web/.env:

OCR_API_URL=http://localhost:8000/api/v1/inference-pipline-ocr
UPLOAD_PRESET=...
CLOUDINARY_API_KEY=...
CLOUDINARY_API_SECRECT=...
CLOUDINARY_CLOUD_NAME=...
CLOUDINARY_URL=...
DATABASE_URL=...
STRIPE_SECRET_KEY=sk_test_dummy
STRIPE_WEBHOOK_SECRET=whsec_dummy
NEXT_PUBLIC_STRIPE_PUBLISHABLE_KEY=pk_test_dummy

4. Access and get API KEY Gemini at Link PASTE API KEY to .env file:

GEMINI_API_KEY=...

3. Install modules detector

cd OCR/detector
git clone https://github.com/aim-uofa/AdelaiDet.git

Add funtion to class Beziers in ./OCR/detector/AdelaiDet/adet/structures/beziers.py:

def __len__(self) -> int:
    """
    Returns:
        int: The number of bezier curves.
    """
    return self.tensor.size(0)

4. Prepare enviroment

Pull docker image

docker pull leduong2408/sign-recognition:v1

5. Run backend:

Turn on CMD and run:

docker-compose up --build

(Note: before run backend, check path-to-project in docker-compose.yml correctly)

6. Run frontend:

Turn on CMD and run:

cd web
npm install
npm run dev