Computr-Vision Handwritten text extractor API

This application is based on hugginface library, in particular microsoft/trocr-base-handwritten.

Demo

Handwritten.Model.-.1.January.2022.mp4

API runs on the 10000 port and have the following endpoints:-

• /isAlive - (get) To verify and check the API running successfullly. sample CURL command

curl -X 'GET' \
  'http://0.0.0.0:10000/isAlive' \
  -H 'accept: application/json'

sample output,

{
  "Message_code": 200,
  "Message": "Alive"
}

• /extractor - (post) Upload the image to to extract the Image written text in and return. sample CURL command,

curl -X 'POST' \
  'http://0.0.0.0:10000/extractor' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@sample.jpg;type=image/jpeg'

sample output

{
  "Message": "File is saved successfully",
  "data": "sample.jpg",
  "results": "Robert Crane"
}

Application is based on following python library stack-

• fastapi
• huggingface
• OpenCV

To deploy the service -

Using docker,

docker build -t ocrExtractorImg .

And run the docker

docker run -p 10000:10000 --name ocrContinaer ocrExtractorImg

Optionally pre-build huggingface models could by volumn mounted

docker run -p 10000:10000 --name ocrContinaer -v <local_path>/models:/models ocrExtractorImg

To use in the local environment, following these steps:

• Install requirements.txt

pip install -r requirements.txt

• Run the server, *This will automatically create the two folder models and temp to which will automatically download the models from the hugginface.

python src/server.py