app.py

import os
import io
import warnings
import easyocr
import numpy as np
from fastapi import FastAPI, UploadFile, File
from fastapi.responses import JSONResponse
from PIL import Image
from pdf2image import convert_from_bytes
from concurrent.futures import ProcessPoolExecutor

# =========================
# EasyOCR config
# =========================
warnings.filterwarnings("ignore", message=".*pin_memory.*")

MODEL_DIR = "/app/.EasyOCR"
USER_NET_DIR = os.path.join(MODEL_DIR, "user_network")

os.makedirs(MODEL_DIR, exist_ok=True)
os.makedirs(USER_NET_DIR, exist_ok=True)

# ✅ preload reader (CPU only for Hugging Face)
reader = easyocr.Reader(
    ['en', 'hi'],
    model_storage_directory=MODEL_DIR,
    user_network_directory=USER_NET_DIR,
    download_enabled=False,
    gpu=False
)

# =========================
# FastAPI app
# =========================
app = FastAPI()


@app.get("/")
async def root():
    return {"message": "OCR API is running on Hugging Face 🚀"}


def run_ocr_on_image(image_bytes: bytes) -> str:
    """OCR worker: takes image bytes, runs OCR, returns text"""
    image = Image.open(io.BytesIO(image_bytes))
    image_np = np.array(image)
    results = reader.readtext(image_np)
    return " ".join([str(text) for _, text, _ in results]).strip()


@app.post("/ocr")
async def ocr(file: UploadFile = File(...)):
    try:
        contents = await file.read()

        if file.filename.lower().endswith(".pdf"):
            # ✅ Convert PDF → images (lower DPI for speed)
            pages = convert_from_bytes(contents, dpi=150)

            # Convert each page to raw bytes (so workers don’t share PIL objects)
            page_bytes = []
            for p in pages:
                buf = io.BytesIO()
                p.save(buf, format="PNG")
                page_bytes.append(buf.getvalue())

            text_results = []
            # ✅ Multiprocessing (parallel CPU workers)
            with ProcessPoolExecutor(max_workers=4) as executor:  # adjust workers
                page_texts = list(executor.map(run_ocr_on_image, page_bytes))

            for i, text in enumerate(page_texts, start=1):
                text_results.append({"page": i, "text": text})

            return JSONResponse(content={"pdf_results": text_results})

        else:
            # ✅ Single image case
            text = run_ocr_on_image(contents)
            return JSONResponse(content={"text": text})

    except Exception as e:
        return JSONResponse(content={"error": str(e)}, status_code=500)