app.py

import gradio as gr
import fitz  # PyMuPDF
from PIL import Image
import requests, json, uuid, os
from io import BytesIO
from concurrent.futures import ThreadPoolExecutor, as_completed
import time


# ================================================================
# FILE NORMALIZER (IMPORTANT for HuggingFace Spaces)
# ================================================================
def resolve_file(file):
    """
    Normalize Gradio file object into a real filesystem file path.
    Handles:
      - dict {name, data}  (HF Spaces)
      - NamedString
      - tempfile object
    """
    # Case 1: HF dict
    if isinstance(file, dict) and "data" in file:
        raw = file["data"]
        fname = file.get("name", f"{uuid.uuid4().hex}.bin")
        path = f"/tmp/{uuid.uuid4().hex}_{os.path.basename(fname)}"
        with open(path, "wb") as f:
            f.write(raw if isinstance(raw, bytes) else raw.read())
        return path

    # Case 2: NamedString (file.name only)
    if hasattr(file, "name") and not hasattr(file, "path"):
        tmp_path = f"/tmp/{uuid.uuid4().hex}_{os.path.basename(file.name)}"
        with open(tmp_path, "wb") as f:
            f.write(open(file.name, "rb").read())
        return tmp_path

    # Case 3: normal tempfile with path
    if hasattr(file, "name"):
        return file.name

    raise ValueError("Unsupported file format:", file)


# ================================================================
# UNIVERSAL RESIZE: max bounds 800×1800, 1800×800, 1200×1200
# ================================================================
def resize_to_max_bounds(img,
                         max_w1=800, max_h1=1800,
                         max_w2=1800, max_h2=800,
                         max_ws=1200, max_hs=1200):
    """Resize image so it stays under max bounds while preserving aspect ratio."""
    w, h = img.size

    bounds = [
        (max_w1, max_h1),
        (max_w2, max_h2),
        (max_ws, max_hs),
    ]

    scale = 1.0
    for max_w, max_h in bounds:
        scale_w = max_w / w
        scale_h = max_h / h
        scale = min(scale, min(scale_w, scale_h))

    if scale >= 1.0:
        return img

    new_size = (int(w * scale), int(h * scale))
    return img.resize(new_size, Image.Resampling.LANCZOS)


# ================================================================
# Preview resize
# ================================================================
def resize_preview(img, max_size=400):
    w, h = img.size
    if max(w, h) <= max_size:
        return img
    scale = max_size / max(w, h)
    return img.resize((int(w * scale), int(h * scale)), Image.Resampling.LANCZOS)


# ================================================================
# Typhoon OCR API call
# ================================================================
def run_typhoon_ocr(img_bytes, api_key, model, task_type,
                    max_tokens, temperature, top_p, repetition_penalty):

    url = "https://api.opentyphoon.ai/v1/ocr"

    files = {"file": ("page.jpg", img_bytes, "image/jpeg")}
    data = {
        "model": model,
        "task_type": task_type,
        "max_tokens": str(max_tokens),
        "temperature": str(temperature),
        "top_p": str(top_p),
        "repetition_penalty": str(repetition_penalty),
    }
    headers = {"Authorization": f"Bearer {api_key}"}

    r = requests.post(url, files=files, data=data, headers=headers)
    if r.status_code != 200:
        return f"❌ Error {r.status_code}\n{r.text}"

    result = r.json()
    texts = []

    for page in result.get("results", []):
        if page.get("success") and page.get("message"):
            content = page["message"]["choices"][0]["message"]["content"]
            try:
                parsed = json.loads(content)
                text = parsed.get("natural_text", content)
            except:
                text = content
            texts.append(text)
        else:
            texts.append(f"❌ Error: {page.get('error')}")

    return "\n\n".join(texts)


# ================================================================
# PDF → Images (PyMuPDF)
# ================================================================
def pdf_to_images_pymupdf(pdf_path, dpi=220):
    doc = fitz.open(pdf_path)
    zoom = dpi / 72
    mat = fitz.Matrix(zoom, zoom)

    images = []
    for page in doc:
        pix = page.get_pixmap(matrix=mat)
        img = Image.open(BytesIO(pix.tobytes("png")))
        images.append(img)
    return images


# ================================================================
# PREVIEW (GRID)
# ================================================================
def preview_files(files):
    previews = []

    for file in files:
        real_path = resolve_file(file)
        fp = real_path.lower()

        if fp.endswith(".pdf"):
            pdf_imgs = pdf_to_images_pymupdf(real_path, dpi=120)
            for img in pdf_imgs:
                img = resize_to_max_bounds(img)
                previews.append(resize_preview(img))
        else:
            img = Image.open(real_path)
            if img.mode == "RGBA":
                img = img.convert("RGB")
            img = resize_to_max_bounds(img)
            previews.append(resize_preview(img))

    return previews


# ================================================================
# OCR 1 PAGE (PARALLEL)
# ================================================================
def ocr_single_page(page_img, label,
                    api_key, model, task_type, max_tokens,
                    temperature, top_p, repetition_penalty):

    buf = BytesIO()
    page_img.convert("RGB").save(buf, format="JPEG")
    buf.seek(0)

    txt = run_typhoon_ocr(
        buf.getvalue(), api_key, model, task_type,
        max_tokens, temperature, top_p, repetition_penalty
    )
    return label, txt


# ================================================================
# MAIN OCR LOGIC
# ================================================================
def extract_text(files,
                 api_key, model, task_type, max_tokens,
                 temperature, top_p, repetition_penalty,
                 progress=gr.Progress(track_tqdm=True)):

    if not files:
        return "❌ No files uploaded.", None

    images_to_ocr = []
    labels = []

    # LOAD FILES
    for file in files:
        real_path = resolve_file(file)
        fp = real_path.lower()

        if fp.endswith(".pdf"):
            pdf_imgs = pdf_to_images_pymupdf(real_path, dpi=220)
            for idx, img in enumerate(pdf_imgs, start=1):
                img = resize_to_max_bounds(img)
                images_to_ocr.append(img)
                labels.append(f"{os.path.basename(real_path)} - Page {idx}")
        else:
            img = Image.open(real_path)
            if img.mode == "RGBA":
                img = img.convert("RGB")
            img = resize_to_max_bounds(img)
            images_to_ocr.append(img)
            labels.append(os.path.basename(real_path))

    total = len(images_to_ocr)
    progress(0.03, desc=f"Loaded {total} pages/images")

    # PARALLEL OCR
    results = {}
    start = time.time()

    with ThreadPoolExecutor(max_workers=4) as ex:
        futures = []
        for img, lbl in zip(images_to_ocr, labels):
            futures.append(
                ex.submit(
                    ocr_single_page, img, lbl,
                    api_key, model, task_type,
                    max_tokens, temperature, top_p, repetition_penalty
                )
            )

        done = 0
        for f in as_completed(futures):
            lbl, txt = f.result()
            results[lbl] = txt

            done += 1
            elapsed = time.time() - start
            eta = (total - done) * (elapsed / max(done, 1))

            progress(done / total,
                     desc=f"OCR {done}/{total} | ETA {eta:.1f}s")

    progress(1, desc="OCR Completed ✔")

    # MERGE RESULT
    merged = ""
    for lbl in sorted(results.keys()):
        merged += f"## {lbl}\n{results[lbl]}\n\n"

    out_path = f"/tmp/ocr_{uuid.uuid4().hex}.txt"
    with open(out_path, "w", encoding="utf-8") as f:
        f.write(merged)

    return merged, out_path


# ================================================================
# UI
# ================================================================
with gr.Blocks() as demo:

    gr.Markdown("""
# 🔍 Typhoon OCR v1.5  
### Multi-file OCR • Parallel Processing • ETA • PDF/Image Support

⚡ **High-speed OCR powered by Typhoon**  
📄 Upload **multiple images or PDFs**  
🚀 Parallel OCR with ETA  
🔍 Auto preview grid for all pages

---

## 🔑 Get Your API Key
👉 https://playground.opentyphoon.ai/settings/api-key  

After logging in, look at the **top-right corner** → you'll see **API Key** menu.  
Click it to generate or copy your key.
""")

    gr.Markdown("### 📘 How to get API Key (step-by-step)")

    with gr.Row():
        gr.Gallery(
            [
                ("ocr_login.png", "1) Login"),
                ("ocr_first.png", "2) Find API Key Menu"),
                ("ocr_getkey.png", "3) Copy Your Key"),
            ],
            columns=3,
            height=250,
            show_label=False,
        )

    file_input = gr.Files(label="Upload images or PDFs", file_count="multiple")

    preview_gallery = gr.Gallery(label="Preview", columns=3, height="auto")
    file_input.change(preview_files, inputs=file_input, outputs=preview_gallery)

    # ADVANCED SETTINGS
    with gr.Accordion("⚙️ Advanced Settings", open=False):
        model_box = gr.Textbox(value="typhoon-ocr", label="Model")
        task_type_box = gr.Textbox(value="v1.5", label="Task Type")
        max_tokens_box = gr.Number(value=16000, label="Max Tokens")
        temperature_box = gr.Number(value=0.1, label="Temperature")
        top_p_box = gr.Number(value=0.6, label="Top-p")
        repetition_penalty_box = gr.Number(value=1.2, label="Repetition Penalty")

    api_key_box = gr.Textbox(label="API Key", type="password")

    run_btn = gr.Button("🚀 Run OCR")

    output_box = gr.Markdown(label="OCR Output")
    download_btn = gr.File(label="Download (.txt)")

    run_btn.click(
        extract_text,
        inputs=[
            file_input,
            api_key_box,
            model_box,
            task_type_box,
            max_tokens_box,
            temperature_box,
            top_p_box,
            repetition_penalty_box,
        ],
        outputs=[output_box, download_btn],
    )

demo.launch()