Revert "fix: Zeilen-Regularisierung im Overlay ueberspringen (generisch fuer gemischte Inhalte)"

This reverts commit b91f799ccf.

admin-lehrer/app/(admin)/ai/ocr-overlay/page.tsx

@@ -216,7 +216,7 @@ export default function OcrOverlayPage() {
       case 3:
         return <StepCrop sessionId={sessionId} onNext={handleNext} />
       case 4:
-        return <StepRowDetection sessionId={sessionId} onNext={handleNext} skipRegularize />
+        return <StepRowDetection sessionId={sessionId} onNext={handleNext} />
       case 5:
         return <StepWordRecognition sessionId={sessionId} onNext={handleNext} goToStep={goToStep} />
       case 6:

admin-lehrer/components/ocr-pipeline/StepRowDetection.tsx

@@ -8,11 +8,9 @@ const KLAUSUR_API = '/klausur-api'
 interface StepRowDetectionProps {
   sessionId: string | null
   onNext: () => void
-  /** Skip word-center regularization (better for mixed-content pages with boxes) */
-  skipRegularize?: boolean
 }
 
-export function StepRowDetection({ sessionId, onNext, skipRegularize = false }: StepRowDetectionProps) {
+export function StepRowDetection({ sessionId, onNext }: StepRowDetectionProps) {
   const [rowResult, setRowResult] = useState<RowResult | null>(null)
   const [detecting, setDetecting] = useState(false)
   const [error, setError] = useState<string | null>(null)
@@ -48,10 +46,7 @@ export function StepRowDetection({ sessionId, onNext, skipRegularize = false }:
     setDetecting(true)
     setError(null)
     try {
-      const rowsUrl = skipRegularize
+      const res = await fetch(`${KLAUSUR_API}/api/v1/ocr-pipeline/sessions/${sessionId}/rows`, {
-        ? `${KLAUSUR_API}/api/v1/ocr-pipeline/sessions/${sessionId}/rows?skip_regularize=true`
-        : `${KLAUSUR_API}/api/v1/ocr-pipeline/sessions/${sessionId}/rows`
-      const res = await fetch(rowsUrl, {
         method: 'POST',
       })
       if (!res.ok) {
@@ -65,7 +60,7 @@ export function StepRowDetection({ sessionId, onNext, skipRegularize = false }:
     } finally {
       setDetecting(false)
     }
-  }, [sessionId, skipRegularize])
+  }, [sessionId])
 
   const handleGroundTruth = useCallback(async (isCorrect: boolean) => {
     if (!sessionId) return

klausur-service/backend/cv_layout.py

@@ -1525,7 +1525,6 @@ def detect_row_geometry(
     word_dicts: List[Dict],
     left_x: int, right_x: int,
     top_y: int, bottom_y: int,
-    skip_regularize: bool = False,
 ) -> List['RowGeometry']:
     """Detect row geometry using horizontal whitespace-gap analysis.
 
@@ -1790,13 +1789,8 @@ def detect_row_geometry(
     # and evenly-spaced rows than the gap-based approach alone.
     # Also detects section breaks (headings, paragraphs) where the pitch
     # exceeds 1.8× the median, and handles each section independently.
-    #
+    rows = _regularize_row_grid(rows, word_dicts, left_x, right_x, top_y,
-    # skip_regularize=True: Keep gap-based rows as-is. Useful for full-page
+                                content_w, content_h, inv)
-    # overlay rendering where mixed content (info boxes, different line
-    # spacings) must preserve original geometry faithfully.
-    if not skip_regularize:
-        rows = _regularize_row_grid(rows, word_dicts, left_x, right_x, top_y,
-                                    content_w, content_h, inv)
 
     type_counts = {}
     for r in rows:

klausur-service/backend/ocr_pipeline_api.py

@@ -1577,7 +1577,7 @@ async def _get_columns_overlay(session_id: str) -> Response:
 # ---------------------------------------------------------------------------
 
 @router.post("/sessions/{session_id}/rows")
-async def detect_rows(session_id: str, skip_regularize: bool = False):
+async def detect_rows(session_id: str):
     """Run row detection on the cropped (or dewarped) image using horizontal gap analysis."""
     if session_id not in _cache:
         await _load_session_to_cache(session_id)
@@ -1686,7 +1686,6 @@ async def detect_rows(session_id: str, skip_regularize: bool = False):
                 combined_h = combined_inv.shape[0]
                 rows = detect_row_geometry(
                     combined_inv, combined_words, left_x, right_x, 0, combined_h,
-                    skip_regularize=skip_regularize,
                 )
 
                 # Remap y-coordinates back to absolute page coords
@@ -1703,12 +1702,10 @@ async def detect_rows(session_id: str, skip_regularize: bool = False):
                     r.y = abs_y
                     r.height = abs_y_end - abs_y
             else:
-                rows = detect_row_geometry(inv, word_dicts, left_x, right_x, top_y, bottom_y,
+                rows = detect_row_geometry(inv, word_dicts, left_x, right_x, top_y, bottom_y)
-                                          skip_regularize=skip_regularize)
         else:
             # No boxes — standard row detection
-            rows = detect_row_geometry(inv, word_dicts, left_x, right_x, top_y, bottom_y,
+            rows = detect_row_geometry(inv, word_dicts, left_x, right_x, top_y, bottom_y)
-                                      skip_regularize=skip_regularize)
 
     duration = time.time() - t0