chore(ocr): add per-step timing logs to identify bottleneck

main.py

@@ -89,6 +89,9 @@ async def ocr_receipt(
             detail=f"Unsupported media type: {content_type}. Upload an image file.",
         )
 
+    import time
+    t0 = time.perf_counter()
+
     raw_bytes = await file.read()
     if not raw_bytes:
         raise HTTPException(status_code=400, detail="Empty file uploaded")
@@ -97,14 +100,20 @@ async def ocr_receipt(
         image = preprocess_image(raw_bytes)
     except (ValueError, FileNotFoundError) as exc:
         raise HTTPException(status_code=400, detail=f"Image preprocessing failed: {exc}")
+    t1 = time.perf_counter()
+    logger.info("TIMING preprocess: %.2fs | image size: %dx%d", t1 - t0, image.shape[1], image.shape[0])
 
     try:
         blocks = _reader.extract(image)
     except Exception as exc:
         logger.exception("OCR extraction failed")
         raise HTTPException(status_code=500, detail=f"OCR failed: {exc}")
+    t2 = time.perf_counter()
+    logger.info("TIMING ocr extract: %.2fs | blocks found: %d", t2 - t1, len(blocks))
 
     result = parse_blocks(blocks)
+    t3 = time.perf_counter()
+    logger.info("TIMING parse: %.2fs | items: %d | total: %.2fs", t3 - t2, len(result.get("line_items", [])), t3 - t0)
 
     if debug:
         result["_raw_blocks"] = blocks

ocr/reader.py

@@ -53,8 +53,17 @@ class PaddleOCRReader:
         img_height = image.shape[0]
         logo_cutoff_y = img_height * 0.10
 
+        import time
+        t0 = time.perf_counter()
         clean = boost_contrast(image)
-        merged = self._run_ocr(_enhance_clahe(clean), logo_confidence)
+        t1 = time.perf_counter()
+        logger.info("TIMING boost_contrast: %.2fs", t1 - t0)
+        clahe = _enhance_clahe(clean)
+        t2 = time.perf_counter()
+        logger.info("TIMING clahe: %.2fs", t2 - t1)
+        merged = self._run_ocr(clahe, logo_confidence)
+        t3 = time.perf_counter()
+        logger.info("TIMING model predict: %.2fs", t3 - t2)
 
         # Apply normal confidence threshold to everything outside the logo zone
         filtered = []