|
|
@@ -192,7 +192,9 @@ class TextFiller:
|
|
|
if rec_item is None:
|
|
|
return "", 0.0
|
|
|
if isinstance(rec_item, tuple) and len(rec_item) >= 2:
|
|
|
- return str(rec_item[0] or "").strip(), float(rec_item[1] or 0.0)
|
|
|
+ txt = str(rec_item[0] or "").strip()
|
|
|
+ sc = float(rec_item[1] or 0.0)
|
|
|
+ return txt, 0.0 if not txt else sc
|
|
|
if isinstance(rec_item, list) and len(rec_item) >= 2:
|
|
|
if isinstance(rec_item[0], (list, tuple, dict)):
|
|
|
texts_list: List[str] = []
|
|
|
@@ -210,11 +212,13 @@ class TextFiller:
|
|
|
return combined, weighted
|
|
|
return combined, sum(scores_list) / len(scores_list)
|
|
|
return "", 0.0
|
|
|
- return str(rec_item[0] or "").strip(), float(rec_item[1] or 0.0)
|
|
|
+ txt = str(rec_item[0] or "").strip()
|
|
|
+ sc = float(rec_item[1] or 0.0)
|
|
|
+ return txt, 0.0 if not txt else sc
|
|
|
if isinstance(rec_item, dict):
|
|
|
txt = str(rec_item.get("text") or rec_item.get("label") or "").strip()
|
|
|
sc = float(rec_item.get("score") or rec_item.get("confidence") or 0.0)
|
|
|
- return txt, sc
|
|
|
+ return txt, 0.0 if not txt else sc
|
|
|
return "", 0.0
|
|
|
|
|
|
def _extract_ocr_batch_results(self, rec_res: Any) -> List[Any]:
|
|
|
@@ -366,8 +370,160 @@ class TextFiller:
|
|
|
return 0.0
|
|
|
|
|
|
return inter_area / ocr_area
|
|
|
-
|
|
|
-
|
|
|
+
|
|
|
+ @staticmethod
|
|
|
+ def _bbox_area(bbox: List[float]) -> float:
|
|
|
+ if not bbox or len(bbox) < 4:
|
|
|
+ return 0.0
|
|
|
+ w = bbox[2] - bbox[0]
|
|
|
+ h = bbox[3] - bbox[1]
|
|
|
+ return max(0.0, w * h)
|
|
|
+
|
|
|
+ @staticmethod
|
|
|
+ def _bbox_from_ocr_original_box(box: Dict[str, Any]) -> List[float]:
|
|
|
+ raw = box.get("original_bbox") or box.get("bbox") or []
|
|
|
+ if not raw:
|
|
|
+ return []
|
|
|
+ if len(raw) >= 4 and not isinstance(raw[0], (list, tuple)):
|
|
|
+ return [float(raw[0]), float(raw[1]), float(raw[2]), float(raw[3])]
|
|
|
+ return CoordinateUtils.poly_to_bbox(raw)
|
|
|
+
|
|
|
+ @staticmethod
|
|
|
+ def _is_bbox_mostly_inside(
|
|
|
+ inner: List[float],
|
|
|
+ outer: List[float],
|
|
|
+ *,
|
|
|
+ inside_ratio: float = 0.7,
|
|
|
+ ) -> bool:
|
|
|
+ """inner 面积的大部分落在 outer 内，且 inner 明显小于 outer。"""
|
|
|
+ if not inner or not outer or len(inner) < 4 or len(outer) < 4:
|
|
|
+ return False
|
|
|
+ inner_area = TextFiller._bbox_area(inner)
|
|
|
+ outer_area = TextFiller._bbox_area(outer)
|
|
|
+ if inner_area <= 0 or outer_area <= 0:
|
|
|
+ return False
|
|
|
+ if inner_area >= outer_area * 0.92:
|
|
|
+ return False
|
|
|
+ inter_x1 = max(inner[0], outer[0])
|
|
|
+ inter_y1 = max(inner[1], outer[1])
|
|
|
+ inter_x2 = min(inner[2], outer[2])
|
|
|
+ inter_y2 = min(inner[3], outer[3])
|
|
|
+ if inter_x2 <= inter_x1 or inter_y2 <= inter_y1:
|
|
|
+ return False
|
|
|
+ inter_area = (inter_x2 - inter_x1) * (inter_y2 - inter_y1)
|
|
|
+ return (inter_area / inner_area) >= inside_ratio
|
|
|
+
|
|
|
+ @staticmethod
|
|
|
+ def _ocr_box_debug_tag(box: Dict[str, Any]) -> str:
|
|
|
+ idx = box.get("paddle_bbox_index")
|
|
|
+ idx_part = f"idx={idx}" if idx is not None else "idx=?"
|
|
|
+ bbox = TextFiller._bbox_from_ocr_original_box(box)
|
|
|
+ if bbox and len(bbox) >= 4:
|
|
|
+ bb = ",".join(f"{v:.0f}" for v in bbox[:4])
|
|
|
+ return f"{idx_part} bbox=[{bb}]"
|
|
|
+ return idx_part
|
|
|
+
|
|
|
+ @staticmethod
|
|
|
+ def _resolve_cell_matched_boxes(
|
|
|
+ matched: List[Tuple[str, float, float, float, float, Dict[str, Any]]],
|
|
|
+ *,
|
|
|
+ cell_idx: Optional[int] = None,
|
|
|
+ y_tolerance: int = 5,
|
|
|
+ inside_ratio: float = 0.7,
|
|
|
+ ) -> Tuple[List[Tuple[str, float, float, float, float, Dict[str, Any]]], bool]:
|
|
|
+ """
|
|
|
+ 处理同格内嵌套 OCR 框：
|
|
|
+ - 大框有字、小框在内：丢弃小框，保留大框文本；
|
|
|
+ - 大框无字、小框在内：丢弃小框，整格 score 置 0（触发二次 OCR）。
|
|
|
+ """
|
|
|
+ if not matched:
|
|
|
+ return matched, False
|
|
|
+
|
|
|
+ matched.sort(key=lambda x: (round(x[1] / y_tolerance), x[2]))
|
|
|
+
|
|
|
+ entries: List[Dict[str, Any]] = []
|
|
|
+ for text, y1, x1, overlap_ratio, score, original_box in matched:
|
|
|
+ bbox = TextFiller._bbox_from_ocr_original_box(original_box)
|
|
|
+ entries.append(
|
|
|
+ {
|
|
|
+ "text": text or "",
|
|
|
+ "y1": y1,
|
|
|
+ "x1": x1,
|
|
|
+ "overlap_ratio": overlap_ratio,
|
|
|
+ "score": score,
|
|
|
+ "original_box": original_box,
|
|
|
+ "bbox": bbox,
|
|
|
+ }
|
|
|
+ )
|
|
|
+
|
|
|
+ remove: set = set()
|
|
|
+ force_zero_score = False
|
|
|
+
|
|
|
+ for i, outer_e in enumerate(entries):
|
|
|
+ outer_bbox = outer_e["bbox"]
|
|
|
+ if not outer_bbox:
|
|
|
+ continue
|
|
|
+ outer_text = (outer_e["text"] or "").strip()
|
|
|
+ for j, inner_e in enumerate(entries):
|
|
|
+ if i == j or j in remove:
|
|
|
+ continue
|
|
|
+ inner_bbox = inner_e["bbox"]
|
|
|
+ if not inner_bbox:
|
|
|
+ continue
|
|
|
+ if not TextFiller._is_bbox_mostly_inside(
|
|
|
+ inner_bbox, outer_bbox, inside_ratio=inside_ratio
|
|
|
+ ):
|
|
|
+ continue
|
|
|
+ inner_text = (inner_e["text"] or "").strip()
|
|
|
+ outer_tag = TextFiller._ocr_box_debug_tag(outer_e["original_box"])
|
|
|
+ inner_tag = TextFiller._ocr_box_debug_tag(inner_e["original_box"])
|
|
|
+ cell_part = f"cell={cell_idx} " if cell_idx is not None else ""
|
|
|
+ if not outer_text:
|
|
|
+ remove.add(j)
|
|
|
+ if inner_text:
|
|
|
+ force_zero_score = True
|
|
|
+ logger.debug(
|
|
|
+ f"{cell_part}嵌套 OCR：空大框套小框，丢弃内框并置 score=0 "
|
|
|
+ f"(outer {outer_tag} text='' | inner {inner_tag} "
|
|
|
+ f"text={inner_text!r} score={inner_e['score']:.3f})"
|
|
|
+ )
|
|
|
+ else:
|
|
|
+ logger.debug(
|
|
|
+ f"{cell_part}嵌套 OCR：空大框套空小框，丢弃内框 "
|
|
|
+ f"(outer {outer_tag} | inner {inner_tag})"
|
|
|
+ )
|
|
|
+ elif inner_text:
|
|
|
+ remove.add(j)
|
|
|
+ logger.debug(
|
|
|
+ f"{cell_part}嵌套 OCR：有字大框套小框，丢弃内框碎片 "
|
|
|
+ f"(outer {outer_tag} text={outer_text!r} | inner {inner_tag} "
|
|
|
+ f"text={inner_text!r} score={inner_e['score']:.3f})"
|
|
|
+ )
|
|
|
+
|
|
|
+ kept = [e for idx, e in enumerate(entries) if idx not in remove]
|
|
|
+ if remove:
|
|
|
+ removed_texts = [
|
|
|
+ (entries[j]["text"] or "").strip()
|
|
|
+ for j in sorted(remove)
|
|
|
+ ]
|
|
|
+ logger.debug(
|
|
|
+ f"{('cell=' + str(cell_idx) + ' ') if cell_idx is not None else ''}"
|
|
|
+ f"嵌套 OCR 汇总: 移除 {len(remove)} 个小框 {removed_texts!r}，"
|
|
|
+ f"保留 {len(kept)} 个框，force_zero_score={force_zero_score}"
|
|
|
+ )
|
|
|
+ resolved = [
|
|
|
+ (
|
|
|
+ e["text"],
|
|
|
+ e["y1"],
|
|
|
+ e["x1"],
|
|
|
+ e["overlap_ratio"],
|
|
|
+ e["score"],
|
|
|
+ e["original_box"],
|
|
|
+ )
|
|
|
+ for e in kept
|
|
|
+ ]
|
|
|
+ return resolved, force_zero_score
|
|
|
+
|
|
|
def fill_text_by_center_point(
|
|
|
self,
|
|
|
bboxes: List[List[float]],
|
|
|
@@ -483,18 +639,21 @@ class TextFiller:
|
|
|
))
|
|
|
|
|
|
if matched:
|
|
|
- # 直接按 y1 和 x1 排序，确保文本顺序正确
|
|
|
- # y_tolerance 用于将相近的 y1 归为同一行（容差范围内视为同一行）
|
|
|
- # 同一行内按 x1 从左到右排序
|
|
|
- y_tolerance = 5
|
|
|
- matched.sort(key=lambda x: (round(x[1] / y_tolerance), x[2])) # 先按 y_group，再按 x1
|
|
|
-
|
|
|
- texts[idx] = "".join([t for t, _, _, _, _, _ in matched])
|
|
|
- # 计算平均置信度
|
|
|
- avg_score = sum([s for _, _, _, _, s, _ in matched]) / len(matched)
|
|
|
- scores[idx] = avg_score
|
|
|
- # 保存匹配到的 OCR boxes
|
|
|
- matched_boxes_list[idx] = [box for _, _, _, _, _, box in matched]
|
|
|
+ matched, force_zero_score = self._resolve_cell_matched_boxes(
|
|
|
+ matched, cell_idx=idx
|
|
|
+ )
|
|
|
+ if matched:
|
|
|
+ texts[idx] = "".join(
|
|
|
+ [(t or "").strip() for t, _, _, _, _, _ in matched]
|
|
|
+ )
|
|
|
+ avg_score = sum(s for _, _, _, _, s, _ in matched) / len(matched)
|
|
|
+ scores[idx] = 0.0 if force_zero_score else avg_score
|
|
|
+ matched_boxes_list[idx] = [
|
|
|
+ box for _, _, _, _, _, box in matched
|
|
|
+ ]
|
|
|
+ else:
|
|
|
+ texts[idx] = ""
|
|
|
+ scores[idx] = 0.0
|
|
|
else:
|
|
|
scores[idx] = 0.0 # 无匹配文本，置信度为0
|
|
|
|
