fix: Zeilen-Regularisierung im Overlay ueberspringen (generisch fuer gemischte Inhalte)
Some checks failed
CI / go-lint (push) Has been skipped
CI / python-lint (push) Has been skipped
CI / nodejs-lint (push) Has been skipped
CI / test-go-school (push) Successful in 49s
CI / test-go-edu-search (push) Successful in 31s
CI / test-python-klausur (push) Failing after 2m21s
CI / test-python-agent-core (push) Successful in 20s
CI / test-nodejs-website (push) Successful in 26s

Seiten mit Info-Boxen (andere Zeilenhoehe) fuehren dazu, dass _regularize_row_grid
die Zeilenpositionen verzerrt. Neuer skip_regularize Parameter nutzt stattdessen
die gap-basierten Zeilen, die der tatsaechlichen Seitengeometrie folgen.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Benjamin Admin
2026-03-11 08:29:06 +01:00
parent 2df2a01a8b
commit b91f799ccf
4 changed files with 23 additions and 9 deletions

View File

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

View File

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

View File

@@ -1525,6 +1525,7 @@ 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.
@@ -1789,8 +1790,13 @@ 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,
content_w, content_h, inv)
#
# skip_regularize=True: Keep gap-based rows as-is. Useful for full-page
# 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:

View File

@@ -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):
async def detect_rows(session_id: str, skip_regularize: bool = False):
"""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)
@@ -1686,6 +1686,7 @@ async def detect_rows(session_id: str):
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
@@ -1702,10 +1703,12 @@ async def detect_rows(session_id: str):
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