Accelerate Tesseract fallback with parallel OCR and timeouts

smoke_signal_tab.py

@@ -22,6 +22,7 @@ Self-improvement loop:
 """
 
 import csv
+import concurrent.futures
 import hashlib
 import json
 import os
@@ -729,6 +730,18 @@ try:
     SS_RENDER_DPI_FALLBACK = max(72, int(os.environ.get("SS_RENDER_DPI_FALLBACK", "200")))
 except Exception:
     SS_RENDER_DPI_FALLBACK = 200
+try:
+    SS_TESSERACT_WORKERS = max(1, int(os.environ.get("SS_TESSERACT_WORKERS", "2")))
+except Exception:
+    SS_TESSERACT_WORKERS = 2
+try:
+    SS_TESSERACT_TIMEOUT_SEC = max(1, int(os.environ.get("SS_TESSERACT_TIMEOUT_SEC", "18")))
+except Exception:
+    SS_TESSERACT_TIMEOUT_SEC = 18
+try:
+    SS_TESSERACT_PSM = max(1, int(os.environ.get("SS_TESSERACT_PSM", "6")))
+except Exception:
+    SS_TESSERACT_PSM = 6
 
 
 def _load_surya_runtime():
@@ -886,14 +899,17 @@ def _run_tesseract_batch(images):
     except Exception as e:
         return [{"regions": [], "confidence": 0.0, "method": f"error-no-tesseract ({e})"} for _ in images]
 
-    outputs = []
-    for img in images:
+    # Prefer parallel image-level OCR with single-threaded internal OpenMP for better CPU utilization.
+    os.environ.setdefault("OMP_THREAD_LIMIT", "1")
+
+    def _ocr_single(img):
         try:
             data = pytesseract.image_to_data(
                 img,
                 lang="eng",
-                config="--oem 1 --psm 6",
+                config=f"--oem 1 --psm {SS_TESSERACT_PSM}",
                 output_type=pytesseract.Output.DICT,
+                timeout=SS_TESSERACT_TIMEOUT_SEC,
             )
             regions = []
             conf_weighted = 0.0
@@ -930,9 +946,28 @@ def _run_tesseract_batch(images):
                 word_count += wc
 
             avg_conf = round(conf_weighted / max(word_count, 1), 4) if regions else 0.0
-            outputs.append({"regions": regions, "confidence": avg_conf, "method": "tesseract"})
+            return {"regions": regions, "confidence": avg_conf, "method": "tesseract"}
+        except RuntimeError as e:
+            return {"regions": [], "confidence": 0.0, "method": f"error-tesseract-timeout ({e})"}
         except Exception as e:
-            outputs.append({"regions": [], "confidence": 0.0, "method": f"error-tesseract ({e})"})
+            return {"regions": [], "confidence": 0.0, "method": f"error-tesseract ({e})"}
+
+    if not images:
+        return []
+
+    max_workers = min(SS_TESSERACT_WORKERS, len(images), max(1, os.cpu_count() or 1))
+    if max_workers <= 1:
+        return [_ocr_single(img) for img in images]
+
+    outputs = [None] * len(images)
+    with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as executor:
+        futures = {executor.submit(_ocr_single, img): idx for idx, img in enumerate(images)}
+        for future in concurrent.futures.as_completed(futures):
+            idx = futures[future]
+            try:
+                outputs[idx] = future.result()
+            except Exception as e:
+                outputs[idx] = {"regions": [], "confidence": 0.0, "method": f"error-tesseract-future ({e})"}
 
     return outputs
 
@@ -976,6 +1011,12 @@ def run_ocr(progress=gr.Progress(track_tqdm=False)) -> tuple:
             log.append(log_line(f"✓ Surya models loaded ({surya['api']})"))
     else:
         log.append(log_line(f"⚠ Surya unavailable ({surya_error}) — using Tesseract fallback for this run"))
+        log.append(
+            log_line(
+                f"ℹ Tesseract fallback config: workers={SS_TESSERACT_WORKERS} timeout={SS_TESSERACT_TIMEOUT_SEC}s "
+                f"psm={SS_TESSERACT_PSM} dpi={SS_RENDER_DPI_FALLBACK}"
+            )
+        )
 
     total_pages_planned = 0
     total_ocr_targets_planned = 0