Health Star Informatics Scence Text Extractor API

Detect and extract regions of text in real world images

Input	Output

How to setup

• Install required dependencies: pip install -r requirements.txt
• Run the production API with gunicorn:

gunicorn --bind 0.0.0.0:8080 wsgi:app --timeout 100 --graceful-timeout 50 --max-requests-jitter 40 --max-requests 40 -w 2  --keep-alive 1

Docker

If you prefer using a docker container. Build a docker images with this command:

docker build . -t scene_text_extractor

And run it with:

docker run -p 8080:8080 -it scene_text_extractor

API

There will be 3 API endpoint running at 0.0.0.0:8080

• \ which can handle full flow from layout analysis to deskew and OCR.
• \layout for layout analysis only.
• \ocr for OCR only.

You can call the API like this:

Here is the sample output from the API:

{
    "prediction": [
        {
            "confidents": 0.9884889125823975,
            "id": 1,
            "location": [
                [
                    238,
                    149
                ],
                [
                    328,
                    123
                ],
                [
                    337,
                    155
                ],
                [
                    248,
                    181
                ]
            ],
            "text": "ROADS"
        },
        {
            "confidents": 0.9965834617614746,
            "id": 2,
            "location": [
                [
                    227,
                    148
                ],
                [
                    257,
                    177
                ],
                [
                    180,
                    257
                ],
                [
                    150,
                    228
                ]
            ],
            "text": "BORDER"
        },
    ],
    "run_time": "3.76",
    "success": true
}