# app.py — TRUST OCR DEMO (Streamlit) — works even if batch_text_detection is missing import os import io import tempfile from typing import List import numpy as np import cv2 from PIL import Image import pypdfium2 import pytesseract # --- set safe dirs before importing streamlit --- safe_home = os.environ.get("HOME") or "/app" os.environ["HOME"] = safe_home os.makedirs(os.path.join(safe_home, ".streamlit"), exist_ok=True) runtime_dir = os.path.join(tempfile.gettempdir(), ".streamlit") os.environ["STREAMLIT_RUNTIME_DIR"] = runtime_dir os.environ["STREAMLIT_CONFIG_DIR"] = os.path.join(safe_home, ".streamlit") os.makedirs(runtime_dir, exist_ok=True) import streamlit as st import streamlit as st # ===== Safe runtime dir for Streamlit/HF cache ===== # runtime_dir = os.path.join(tempfile.gettempdir(), ".streamlit") # os.environ["STREAMLIT_RUNTIME_DIR"] = runtime_dir # os.makedirs(runtime_dir, exist_ok=True) # ===== Try to import Surya APIs ===== DET_AVAILABLE = True try: from surya.detection import batch_text_detection except Exception: DET_AVAILABLE = False from surya.layout import batch_layout_detection # may still import; we’ll gate usage by DET_AVAILABLE # Detection model loaders: segformer (newer) vs model (older) try: from surya.model.detection.segformer import load_model as load_det_model, load_processor as load_det_processor except Exception: from surya.model.detection.model import load_model as load_det_model, load_processor as load_det_processor from surya.model.recognition.model import load_model as load_rec_model from surya.model.recognition.processor import load_processor as load_rec_processor from surya.model.ordering.model import load_model as load_order_model from surya.model.ordering.processor import load_processor as load_order_processor from surya.ordering import batch_ordering from surya.ocr import run_ocr from surya.postprocessing.heatmap import draw_polys_on_image from surya.postprocessing.text import draw_text_on_image from surya.languages import CODE_TO_LANGUAGE from surya.input.langs import replace_lang_with_code from surya.schema import OCRResult, TextDetectionResult, LayoutResult, OrderResult # ===================== Helper Functions ===================== def remove_border(image_path: str, output_path: str) -> np.ndarray: """Remove outer border & deskew (perspective) if a rectangular contour is found.""" image = cv2.imread(image_path) if image is None: raise ValueError(f"Cannot read image: {image_path}") gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) _, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU) contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) if not contours: cv2.imwrite(output_path, image) return image max_contour = max(contours, key=cv2.contourArea) epsilon = 0.02 * cv2.arcLength(max_contour, True) approx = cv2.approxPolyDP(max_contour, epsilon, True) if len(approx) == 4: pts = approx.reshape(4, 2).astype("float32") rect = np.zeros((4, 2), dtype="float32") s = pts.sum(axis=1) rect[0] = pts[np.argmin(s)] # tl rect[2] = pts[np.argmax(s)] # br diff = np.diff(pts, axis=1) rect[1] = pts[np.argmin(diff)] # tr rect[3] = pts[np.argmax(diff)] # bl (tl, tr, br, bl) = rect widthA = np.linalg.norm(br - bl) widthB = np.linalg.norm(tr - tl) maxWidth = max(int(widthA), int(widthB)) heightA = np.linalg.norm(tr - br) heightB = np.linalg.norm(tl - bl) maxHeight = max(int(heightA), int(heightB)) dst = np.array([[0, 0], [maxWidth - 1, 0], [maxWidth - 1, maxHeight - 1], [0, maxHeight - 1]], dtype="float32") M = cv2.getPerspectiveTransform(rect, dst) cropped = cv2.warpPerspective(image, M, (maxWidth, maxHeight)) cv2.imwrite(output_path, cropped) return cropped else: cv2.imwrite(output_path, image) return image def open_pdf(pdf_file) -> pypdfium2.PdfDocument: stream = io.BytesIO(pdf_file.getvalue()) return pypdfium2.PdfDocument(stream) @st.cache_data(show_spinner=False) def get_page_image(pdf_file, page_num: int, dpi: int = 96) -> Image.Image: doc = open_pdf(pdf_file) renderer = doc.render(pypdfium2.PdfBitmap.to_pil, page_indices=[page_num - 1], scale=dpi / 72) png = list(renderer)[0] return png.convert("RGB") @st.cache_data(show_spinner=False) def page_count(pdf_file) -> int: doc = open_pdf(pdf_file) return len(doc) # ===================== Streamlit UI ===================== st.set_page_config(page_title="TRUST OCR DEMO", layout="wide") st.markdown("# TRUST OCR DEMO") if not DET_AVAILABLE: st.warning("⚠️ ماژول تشخیص متن Surya در این محیط در دسترس نیست. OCR کامل کار می‌کند، اما دکمه‌های Detection/Layout/Order غیرفعال شده‌اند. برای فعال‌سازی آن‌ها، Surya را به نسخهٔ سازگار پین کنید (راهنما پایین صفحه).") # Sidebar controls in_file = st.sidebar.file_uploader("فایل PDF یا عکس :", type=["pdf", "png", "jpg", "jpeg", "gif", "webp"]) languages = st.sidebar.multiselect( "زبان‌ها (Languages)", sorted(list(CODE_TO_LANGUAGE.values())), default=["Persian"], max_selections=4 ) auto_rotate = st.sidebar.toggle("چرخش خودکار (Tesseract OSD)", value=True) auto_border = st.sidebar.toggle("حذف قاب/کادر تصویر ورودی", value=True) # Buttons (disable some if detection missing) text_det_btn = st.sidebar.button("تشخیص متن (Detection)", disabled=not DET_AVAILABLE) layout_det_btn = st.sidebar.button("آنالیز صفحه (Layout)", disabled=not DET_AVAILABLE) order_det_btn = st.sidebar.button("ترتیب خوانش (Reading Order)", disabled=not DET_AVAILABLE) text_rec_btn = st.sidebar.button("تبدیل به متن (Recognition)") if in_file is None: st.info("یک فایل PDF/عکس از سایدبار انتخاب کنید. | Please upload a file to begin.") st.stop() filetype = in_file.type # Two-column layout (left: outputs / right: input image) col2, col1 = st.columns([.5, .5]) # ===================== Load Models (cached) ===================== @st.cache_resource(show_spinner=True) def load_det_cached(): return load_det_model(checkpoint="vikp/surya_det2"), load_det_processor(checkpoint="vikp/surya_det2") @st.cache_resource(show_spinner=True) def load_rec_cached(): return load_rec_model(checkpoint="MohammadReza-Halakoo/TrustOCR"), \ load_rec_processor(checkpoint="MohammadReza-Halakoo/TrustOCR") @st.cache_resource(show_spinner=True) def load_layout_cached(): return load_det_model(checkpoint="vikp/surya_layout2"), load_det_processor(checkpoint="vikp/surya_layout2") @st.cache_resource(show_spinner=True) def load_order_cached(): return load_order_model(checkpoint="vikp/surya_order"), load_order_processor(checkpoint="vikp/surya_order") # recognition models are enough for run_ocr; detection/layout/order models used only if DET_AVAILABLE rec_model, rec_processor = load_rec_cached() if DET_AVAILABLE: det_model, det_processor = load_det_cached() layout_model, layout_processor = load_layout_cached() order_model, order_processor = load_order_cached() else: det_model = det_processor = layout_model = layout_processor = order_model = order_processor = None # ===================== High-level Ops ===================== def _apply_auto_rotate(pil_img: Image.Image) -> Image.Image: """Auto-rotate using Tesseract OSD if enabled.""" if not auto_rotate: return pil_img try: osd = pytesseract.image_to_osd(pil_img, output_type=pytesseract.Output.DICT) angle = int(osd.get("rotate", 0)) # 0/90/180/270 if angle and angle % 360 != 0: return pil_img.rotate(-angle, expand=True) return pil_img except Exception as e: st.warning(f"OSD rotation failed, continuing without rotation. Error: {e}") return pil_img def text_detection(pil_img: Image.Image): pred: TextDetectionResult = batch_text_detection([pil_img], det_model, det_processor)[0] polygons = [p.polygon for p in pred.bboxes] det_img = draw_polys_on_image(polygons, pil_img.copy()) return det_img, pred def layout_detection(pil_img: Image.Image): _, det_pred = text_detection(pil_img) pred: LayoutResult = batch_layout_detection([pil_img], layout_model, layout_processor, [det_pred])[0] polygons = [p.polygon for p in pred.bboxes] labels = [p.label for p in pred.bboxes] layout_img = draw_polys_on_image(polygons, pil_img.copy(), labels=labels, label_font_size=40) return layout_img, pred def order_detection(pil_img: Image.Image): _, layout_pred = layout_detection(pil_img) bboxes = [l.bbox for l in layout_pred.bboxes] pred: OrderResult = batch_ordering([pil_img], [bboxes], order_model, order_processor)[0] polys = [l.polygon for l in pred.bboxes] positions = [str(l.position) for l in pred.bboxes] order_img = draw_polys_on_image(polys, pil_img.copy(), labels=positions, label_font_size=40) return order_img, pred def ocr_page(pil_img: Image.Image, langs: List[str]): """Full-page OCR using Surya run_ocr — works without detection import.""" langs = list(langs) if langs else ["Persian"] replace_lang_with_code(langs) # in-place # If detection models are loaded, pass them; else, let run_ocr use its internal defaults args = [pil_img], [langs] if det_model and det_processor and rec_model and rec_processor: img_pred: OCRResult = run_ocr([pil_img], [langs], det_model, det_processor, rec_model, rec_processor)[0] else: img_pred: OCRResult = run_ocr([pil_img], [langs])[0] bboxes = [l.bbox for l in img_pred.text_lines] text = [l.text for l in img_pred.text_lines] rec_img = draw_text_on_image(bboxes, text, pil_img.size, langs, has_math="_math" in langs) return rec_img, img_pred # ===================== Input Handling ===================== if "pdf" in filetype: try: pg_cnt = page_count(in_file) except Exception as e: st.error(f"خواندن PDF ناموفق بود: {e}") st.stop() page_number = st.sidebar.number_input("صفحه:", min_value=1, value=1, max_value=pg_cnt) pil_image = get_page_image(in_file, page_number) else: bytes_data = in_file.getvalue() temp_dir = "temp_files" os.makedirs(temp_dir, exist_ok=True) file_path = os.path.join(temp_dir, in_file.name) with open(file_path, "wb") as f: f.write(bytes_data) out_file = os.path.splitext(file_path)[0] + "-1.JPG" try: if auto_border: _ = remove_border(file_path, out_file) pil_image = Image.open(out_file).convert("RGB") else: pil_image = Image.open(file_path).convert("RGB") except Exception as e: st.warning(f"حذف قاب/بازخوانی تصویر با خطا مواجه شد؛ تصویر اصلی استفاده می‌شود. Error: {e}") pil_image = Image.open(file_path).convert("RGB") # Auto-rotate if enabled pil_image = _apply_auto_rotate(pil_image) # ===================== Buttons Logic ===================== with col1: if text_det_btn and DET_AVAILABLE: try: det_img, det_pred = text_detection(pil_image) st.image(det_img, caption="تشخیص متن (Detection)", use_column_width=True) except Exception as e: st.error(f"خطا در تشخیص متن: {e}") if layout_det_btn and DET_AVAILABLE: try: layout_img, layout_pred = layout_detection(pil_image) st.image(layout_img, caption="آنالیز صفحه (Layout)", use_column_width=True) except Exception as e: st.error(f"خطا در آنالیز صفحه: {e}") if order_det_btn and DET_AVAILABLE: try: order_img, order_pred = order_detection(pil_image) st.image(order_img, caption="ترتیب خوانش (Reading Order)", use_column_width=True) except Exception as e: st.error(f"خطا در ترتیب خوانش: {e}") if text_rec_btn: try: rec_img, ocr_pred = ocr_page(pil_image, languages) text_tab, json_tab = st.tabs(["متن صفحه | Page Text", "JSON"]) with text_tab: st.text("\n".join([p.text for p in ocr_pred.text_lines])) with json_tab: st.json(ocr_pred.model_dump(), expanded=False) except Exception as e: st.error(f"خطا در بازشناسی متن (Recognition): {e}") with col2: st.image(pil_image, caption="تصویر ورودی | Input Preview", use_column_width=True)