Speed up fallback OCR rendering and keep progress moving

smoke_signal_tab.py

@@ -726,9 +726,9 @@ try:
 except Exception:
     SS_RENDER_DPI_SURYA = 300
 try:
-    SS_RENDER_DPI_FALLBACK = max(72, int(os.environ.get("SS_RENDER_DPI_FALLBACK", "220")))
+    SS_RENDER_DPI_FALLBACK = max(72, int(os.environ.get("SS_RENDER_DPI_FALLBACK", "200")))
 except Exception:
-    SS_RENDER_DPI_FALLBACK = 220
+    SS_RENDER_DPI_FALLBACK = 200
 
 
 def _load_surya_runtime():
@@ -978,6 +978,7 @@ def run_ocr(progress=gr.Progress(track_tqdm=False)) -> tuple:
         log.append(log_line(f"⚠ Surya unavailable ({surya_error}) — using Tesseract fallback for this run"))
 
     total_pages_planned = 0
+    total_ocr_targets_planned = 0
     for _, row in eligible.iterrows():
         profile_path = PROFILES_DIR / f"{row['book_id']}_page_profile.json"
         if not profile_path.exists():
@@ -985,13 +986,16 @@ def run_ocr(progress=gr.Progress(track_tqdm=False)) -> tuple:
         try:
             with open(profile_path, encoding="utf-8") as f:
                 profile = json.load(f)
-            total_pages_planned += len(profile.get("pages", []))
+            pages = profile.get("pages", [])
+            total_pages_planned += len(pages)
+            total_ocr_targets_planned += sum(1 for p in pages if p.get("route") != "embedded_text")
         except Exception:
             continue
     total_pages_planned = max(total_pages_planned, 1)
-    processed_pages = 0
+    total_ocr_targets_planned = max(total_ocr_targets_planned, 1)
     rendered_pages = 0
-    total_work_units = max(total_pages_planned * 2, 1)
+    ocr_processed_pages = 0
+    total_work_units = max(total_pages_planned + total_ocr_targets_planned, 1)
 
     for _, row in eligible.iterrows():
         book_id = row["book_id"]
@@ -1035,14 +1039,37 @@ def run_ocr(progress=gr.Progress(track_tqdm=False)) -> tuple:
             render_path = None
             if doc is not None:
                 try:
-                    page = doc[page_num - 1]
-                    pix = page.get_pixmap(dpi=render_dpi, alpha=False)
-                    render_dir = RENDERS_DIR / book_id
-                    render_dir.mkdir(exist_ok=True)
-                    render_path = render_dir / f"{book_id}_page_{page_num:04d}_{render_dpi}dpi.png"
-                    pix.save(str(render_path))
-                    page_data["render_path"] = str(render_path.relative_to(SS_ROOT))
-                    page_data["render_dpi"] = render_dpi
+                    existing_rel = page_data.get("render_path")
+                    existing_dpi = int(page_data.get("render_dpi", 0) or 0)
+                    existing_abs = (SS_ROOT / existing_rel) if existing_rel else None
+                    if existing_abs and existing_abs.exists() and existing_dpi == render_dpi:
+                        render_path = existing_abs
+                    else:
+                        page = doc[page_num - 1]
+                        render_start = time.time()
+                        if surya is None:
+                            pix = page.get_pixmap(
+                                dpi=render_dpi,
+                                alpha=False,
+                                annots=False,
+                                colorspace=fitz.csGRAY,
+                            )
+                        else:
+                            pix = page.get_pixmap(
+                                dpi=render_dpi,
+                                alpha=False,
+                                annots=False,
+                                colorspace=fitz.csRGB,
+                            )
+                        render_dir = RENDERS_DIR / book_id
+                        render_dir.mkdir(exist_ok=True)
+                        render_path = render_dir / f"{book_id}_page_{page_num:04d}_{render_dpi}dpi.png"
+                        pix.save(str(render_path))
+                        render_secs = round(time.time() - render_start, 2)
+                        if render_secs >= 4.0:
+                            log.append(log_line(f"  ⚠ {book_id} p{page_num}: slow render {render_secs}s at {render_dpi}dpi"))
+                        page_data["render_path"] = str(render_path.relative_to(SS_ROOT))
+                        page_data["render_dpi"] = render_dpi
                 except Exception as e:
                     log.append(log_line(f"  ⚠ {book_id} p{page_num}: render failed ({e})"))
 
@@ -1111,10 +1138,20 @@ def run_ocr(progress=gr.Progress(track_tqdm=False)) -> tuple:
                                 "confidence": conf,
                                 "method": "surya",
                             }
+                            ocr_processed_pages += 1
+                            progress(
+                                (total_pages_planned + ocr_processed_pages, total_work_units),
+                                desc=f"OCR {book_id} p{item['page_num']}",
+                            )
                     else:
                         fallback_preds = _run_tesseract_batch(batch_images)
                         for item, pred in zip(batch_items_with_images, fallback_preds):
                             ocr_lookup[item["page_num"]] = pred
+                            ocr_processed_pages += 1
+                            progress(
+                                (total_pages_planned + ocr_processed_pages, total_work_units),
+                                desc=f"Tesseract {book_id} p{item['page_num']}",
+                            )
                 except Exception as e:
                     # If Surya batch fails, try Tesseract for this batch before giving up.
                     if surya is not None:
@@ -1122,6 +1159,11 @@ def run_ocr(progress=gr.Progress(track_tqdm=False)) -> tuple:
                         fallback_preds = _run_tesseract_batch(batch_images)
                         for item, pred in zip(batch_items_with_images, fallback_preds):
                             ocr_lookup[item["page_num"]] = pred
+                            ocr_processed_pages += 1
+                            progress(
+                                (total_pages_planned + ocr_processed_pages, total_work_units),
+                                desc=f"Tesseract retry {book_id} p{item['page_num']}",
+                            )
                     else:
                         for item in batch_items_with_images:
                             ocr_lookup[item["page_num"]] = {
@@ -1142,11 +1184,6 @@ def run_ocr(progress=gr.Progress(track_tqdm=False)) -> tuple:
             page_num = page_data["page_number"]
             route = page_data["route"]
 
-            progress(
-                (total_pages_planned + processed_pages, total_work_units),
-                desc=f"OCR {book_id} p{page_num}/{len(profile.get('pages', []))}",
-            )
-
             if route == "embedded_text":
                 try:
                     if doc is None:
@@ -1211,8 +1248,6 @@ def run_ocr(progress=gr.Progress(track_tqdm=False)) -> tuple:
                     "reason_code": "",
                 })
 
-            processed_pages += 1
-
         if doc is not None:
             try:
                 doc.close()