Identity Card Detector

You can check the application from the website below,

https://identity-detector.herokuapp.com

Install

Install dependencies by using pip,

$ pip3.9 install -r requirements.txt

Training

You need to download dataset with using download script,

$ python3 download.py ./default.ini # or custom ini file

Then train your model as follows.

Please keep in mind, if you've a saved model in the checkpoint path, the command below continues the training.

$ python3 train.py ./default.ini # or custom ini file

Please notice that, you need to change the path variables in the default config file named as default.ini.

You can follow your model on the tensorboard,

$ tensorboard --logdir /tmp/tensorboard-logs

Please notice that, If you change logdir variable under the model section in the default config, you need to change the path to /tmp/tensorboard-logs.

Prediction

You can download model here.

Then, predict with using folder or only one file as follows,

$ python3 predict.py --model /path/to/model --path /path/to/folder # or /path/to/file

Here is the example response for a folder,

[
    {
        "name": "identity_card.jpg",
        "prediction": "document",
        "probability": 99.99996423721313
    },
    {
        "name": "non_document.jpg",
        "prediction": "non_document",
        "probability": 100.0
    },
    {
        "name": "selfie.jpg",
        "prediction": "selfie",
        "probability": 99.23766255378723
    }
]

Or you can use your camera,

$ python3 stream.py --model /path/to/model

Examples