app.py

import gradio as gr
from transformers import DonutProcessor, VisionEncoderDecoderModel


processor = DonutProcessor.from_pretrained("nielsr/donut-demo")
model = VisionEncoderDecoderModel.from_pretrained("nielsr/donut-demo")


def donut(input_img):
    # prepare encoder inputs
    pixel_values = processor(sample["image"].convert("RGB"), return_tensors="pt").pixel_values
    pixel_values = pixel_values.to(device)
    # prepare decoder inputs
    task_prompt = "<s_cord-v2>"
    decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
    decoder_input_ids = decoder_input_ids.to(device)

    # autoregressively generate sequence
    model = model.to(device)
    return model.generate(
        pixel_values,
        decoder_input_ids=decoder_input_ids,
        max_length=model.decoder.config.max_position_embeddings,
        early_stopping=True,
        pad_token_id=processor.tokenizer.pad_token_id,
        eos_token_id=processor.tokenizer.eos_token_id,
        use_cache=True,
        num_beams=1,
        bad_words_ids=[[processor.tokenizer.unk_token_id]],
        return_dict_in_generate=True,
    )


def parse_json(outputs):
    seq = processor.batch_decode(outputs.sequences)[0]
    seq = seq.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
    seq = re.sub(r"<.*?>", "", seq, count=1).strip()  # remove first task start token

    return processor.token2json(seq)


def predict(input_img):
    outputs = donut(input_img)
    result = parse_json(outputs)

    return result


gradio_app = gr.Interface(
    predict,
    inputs=gr.Image(label="Upload gambar dokumen", sources=['upload', 'webcam'], type="pil"),
    outputs=[gr.JSON(label="Hasil")],
    title="OCR Dokumen Identitas Indonesia",
    description="Ekstraksi gambar KTP, SIM, Paspor, KK, dan NPWP menjadi data teks tersturktur",
)


if __name__ == "__main__":
    gradio_app.launch(share=True)