app.py

import json, base64
from io import BytesIO
from PIL import Image
import gradio as gr
from inference import OcrReorderPipeline
from transformers import (
    AutoProcessor,
    LayoutLMv3Model,
    AutoTokenizer
)
import torch

# 1) Load from your model repo, pointing at the `preprocessor/` folder
repo = "Uddipan107/ocr-layoutlmv3-base-t5-small"
model     = LayoutLMv3Model.from_pretrained(repo)
tokenizer = AutoTokenizer.from_pretrained(repo, subfolder="preprocessor")
processor = AutoProcessor.from_pretrained(repo, subfolder="preprocessor", apply_ocr=False)

# 2) Instantiate your pipeline
pipe = OcrReorderPipeline(model, tokenizer, processor, device=0)

def infer(image, words_json, boxes_json):
    words = json.loads(words_json)
    boxes = json.loads(boxes_json)

    # Encode PIL image → PNG → base64
    buf = BytesIO()
    image.save(buf, format="PNG")
    b64 = base64.b64encode(buf.getvalue()).decode()

    # Run your custom pipeline and return the first (only) output string
    return pipe(b64, words, boxes)[0]

# 3) Gradio UI
demo = gr.Interface(
    fn=infer,
    inputs=[
      gr.Image(type="pil", label="Image"),
      gr.Textbox(label="Words (JSON list)"),
      gr.Textbox(label="Boxes (JSON list)")
    ],
    outputs="text",
    title="OCR Reorder Pipeline"
)

if __name__ == "__main__":
    demo.launch()