ID Card Segmentation with U-Net

Performing semantic segmentation on ID cards using U-Net implemented in Tensorflow 2.10.0

Example Usage

Run python example.py -i <path-to-image> -o <path-to-output-file> to generate a segmented ID card image.

See example.py for more details.

Implementation

See the U-Net Semantic Segmentation notebook for comprehensive details regarding model implementation, training, inference and a complete pipeline example.

Model

You can use the pre-trained model best_model.h5. This model achieved 96.5% mean IoU on the test set.

References

1. Github: ML_IDCard_Segmentation-TF-Keras

2. Dataset: MIDV-500

3. Perspective transformation from How to Build a Kick-Ass Mobile Document Scanner by PyimageSearch

4. Github: Imutils