feat: implement table cell boxes extraction with SLANeXt
Phase 1-3 implementation of extract-table-cell-boxes proposal: - Add enable_table_cell_boxes_extraction config option - Implement lazy-loaded SLANeXt model caching in PPStructureEnhanced - Add _extract_cell_boxes_with_slanet() method for direct model invocation - Supplement PPStructureV3 table processing with SLANeXt cell boxes - Add _compute_table_grid_from_cell_boxes() for column width calculation - Modify draw_table_region() to use cell_boxes for accurate layout Key features: - Auto-detect table type (wired/wireless) using PP-LCNet classifier - Convert 8-point polygon bbox to 4-point rectangle - Graceful fallback to equal distribution when cell_boxes unavailable - Proper coordinate transformation with scaling support 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
egg
@@ -1605,6 +1605,120 @@ class PDFGeneratorService:
|
||||
except Exception as e:
|
||||
logger.warning(f"Failed to draw text region '{text[:20]}...': {e}")
|
||||
|
||||
def _compute_table_grid_from_cell_boxes(
|
||||
self,
|
||||
cell_boxes: List[List[float]],
|
||||
table_bbox: List[float],
|
||||
num_rows: int,
|
||||
num_cols: int
|
||||
) -> Tuple[Optional[List[float]], Optional[List[float]]]:
|
||||
"""
|
||||
Compute column widths and row heights from cell bounding boxes.
|
||||
|
||||
This uses the cell boxes extracted by SLANeXt to calculate the actual
|
||||
column widths and row heights, which provides more accurate table rendering
|
||||
than uniform distribution.
|
||||
|
||||
Args:
|
||||
cell_boxes: List of cell bboxes [[x1,y1,x2,y2], ...]
|
||||
table_bbox: Table bounding box [x1,y1,x2,y2]
|
||||
num_rows: Number of rows in the table
|
||||
num_cols: Number of columns in the table
|
||||
|
||||
Returns:
|
||||
Tuple of (col_widths, row_heights) or (None, None) if calculation fails
|
||||
"""
|
||||
if not cell_boxes or len(cell_boxes) < 2:
|
||||
return None, None
|
||||
|
||||
try:
|
||||
table_x1, table_y1, table_x2, table_y2 = table_bbox
|
||||
table_width = table_x2 - table_x1
|
||||
table_height = table_y2 - table_y1
|
||||
|
||||
# Collect all unique X and Y boundaries from cell boxes
|
||||
x_boundaries = set()
|
||||
y_boundaries = set()
|
||||
|
||||
for box in cell_boxes:
|
||||
if len(box) >= 4:
|
||||
x1, y1, x2, y2 = box[:4]
|
||||
# Convert to relative coordinates within table
|
||||
x_boundaries.add(x1 - table_x1)
|
||||
x_boundaries.add(x2 - table_x1)
|
||||
y_boundaries.add(y1 - table_y1)
|
||||
y_boundaries.add(y2 - table_y1)
|
||||
|
||||
# Sort boundaries
|
||||
x_boundaries = sorted(x_boundaries)
|
||||
y_boundaries = sorted(y_boundaries)
|
||||
|
||||
# Ensure we have boundaries at table edges
|
||||
if x_boundaries and x_boundaries[0] > 5:
|
||||
x_boundaries.insert(0, 0)
|
||||
if x_boundaries and x_boundaries[-1] < table_width - 5:
|
||||
x_boundaries.append(table_width)
|
||||
|
||||
if y_boundaries and y_boundaries[0] > 5:
|
||||
y_boundaries.insert(0, 0)
|
||||
if y_boundaries and y_boundaries[-1] < table_height - 5:
|
||||
y_boundaries.append(table_height)
|
||||
|
||||
# Calculate column widths from X boundaries
|
||||
# Merge boundaries that are too close (< 5px)
|
||||
merged_x = [x_boundaries[0]] if x_boundaries else []
|
||||
for x in x_boundaries[1:]:
|
||||
if x - merged_x[-1] > 5:
|
||||
merged_x.append(x)
|
||||
x_boundaries = merged_x
|
||||
|
||||
# Calculate row heights from Y boundaries
|
||||
merged_y = [y_boundaries[0]] if y_boundaries else []
|
||||
for y in y_boundaries[1:]:
|
||||
if y - merged_y[-1] > 5:
|
||||
merged_y.append(y)
|
||||
y_boundaries = merged_y
|
||||
|
||||
# Calculate widths and heights
|
||||
col_widths = []
|
||||
for i in range(len(x_boundaries) - 1):
|
||||
col_widths.append(x_boundaries[i + 1] - x_boundaries[i])
|
||||
|
||||
row_heights = []
|
||||
for i in range(len(y_boundaries) - 1):
|
||||
row_heights.append(y_boundaries[i + 1] - y_boundaries[i])
|
||||
|
||||
# Validate: number of columns/rows should match expected
|
||||
if len(col_widths) == num_cols and len(row_heights) == num_rows:
|
||||
logger.info(f"[TABLE] Cell boxes grid: {num_cols} cols, {num_rows} rows")
|
||||
logger.debug(f"[TABLE] Col widths from cell_boxes: {[f'{w:.1f}' for w in col_widths]}")
|
||||
logger.debug(f"[TABLE] Row heights from cell_boxes: {[f'{h:.1f}' for h in row_heights]}")
|
||||
return col_widths, row_heights
|
||||
else:
|
||||
# Grid doesn't match, might be due to merged cells
|
||||
logger.debug(
|
||||
f"[TABLE] Cell boxes grid mismatch: "
|
||||
f"got {len(col_widths)}x{len(row_heights)}, expected {num_cols}x{num_rows}"
|
||||
)
|
||||
# Still return the widths/heights if counts are close
|
||||
if abs(len(col_widths) - num_cols) <= 1 and abs(len(row_heights) - num_rows) <= 1:
|
||||
# Adjust to match expected count
|
||||
while len(col_widths) < num_cols:
|
||||
col_widths.append(col_widths[-1] if col_widths else table_width / num_cols)
|
||||
while len(col_widths) > num_cols:
|
||||
col_widths.pop()
|
||||
while len(row_heights) < num_rows:
|
||||
row_heights.append(row_heights[-1] if row_heights else table_height / num_rows)
|
||||
while len(row_heights) > num_rows:
|
||||
row_heights.pop()
|
||||
return col_widths, row_heights
|
||||
|
||||
return None, None
|
||||
|
||||
except Exception as e:
|
||||
logger.warning(f"[TABLE] Failed to compute grid from cell boxes: {e}")
|
||||
return None, None
|
||||
|
||||
def draw_table_region(
|
||||
self,
|
||||
pdf_canvas: canvas.Canvas,
|
||||
@@ -1765,8 +1879,36 @@ class PDFGeneratorService:
|
||||
|
||||
reportlab_data.append(row_data)
|
||||
|
||||
# Calculate column widths (equal distribution)
|
||||
col_widths = [table_width / max_cols] * max_cols
|
||||
# Calculate column widths and row heights
|
||||
# First, try to use cell_boxes if available for more accurate layout
|
||||
cell_boxes = table_element.get('cell_boxes')
|
||||
raw_table_bbox = [ocr_x_left_raw, ocr_y_top_raw, ocr_x_right_raw, ocr_y_bottom_raw]
|
||||
|
||||
computed_col_widths = None
|
||||
computed_row_heights = None
|
||||
|
||||
if cell_boxes:
|
||||
cell_boxes_source = table_element.get('cell_boxes_source', 'unknown')
|
||||
logger.info(f"[TABLE] Using {len(cell_boxes)} cell boxes from {cell_boxes_source}")
|
||||
computed_col_widths, computed_row_heights = self._compute_table_grid_from_cell_boxes(
|
||||
cell_boxes, raw_table_bbox, num_rows, max_cols
|
||||
)
|
||||
|
||||
# Use computed widths if available, otherwise fall back to equal distribution
|
||||
if computed_col_widths:
|
||||
# Scale col_widths to PDF coordinates
|
||||
col_widths = [w * scale_w for w in computed_col_widths]
|
||||
logger.info(f"[TABLE] Using cell_boxes col widths (scaled)")
|
||||
else:
|
||||
col_widths = [table_width / max_cols] * max_cols
|
||||
logger.info(f"[TABLE] Using equal distribution col widths")
|
||||
|
||||
# Row heights are used optionally (ReportLab can auto-size)
|
||||
row_heights = None
|
||||
if computed_row_heights:
|
||||
# Scale row_heights to PDF coordinates
|
||||
row_heights = [h * scale_h for h in computed_row_heights]
|
||||
logger.debug(f"[TABLE] Cell_boxes row heights available (scaled)")
|
||||
|
||||
# Create ReportLab Table
|
||||
# Use smaller font to fit content with auto-wrap
|
||||
@@ -1790,7 +1932,11 @@ class PDFGeneratorService:
|
||||
escaped_text = cell_text.replace('&', '&').replace('<', '<').replace('>', '>')
|
||||
reportlab_data[row_idx][col_idx] = Paragraph(escaped_text, cell_style)
|
||||
|
||||
# Create table WITHOUT fixed row heights - let it auto-size based on content
|
||||
# Create table with computed col widths
|
||||
# Note: We don't use row_heights even when available from cell_boxes because:
|
||||
# 1. ReportLab's auto-sizing handles content overflow better
|
||||
# 2. Fixed heights can cause text clipping when content exceeds cell size
|
||||
# 3. The col_widths from cell_boxes provide the main layout benefit
|
||||
table = Table(reportlab_data, colWidths=col_widths)
|
||||
|
||||
# Apply table style
|
||||
|
||||
Reference in New Issue
Block a user