Skip to content

GitLab

Commit 1906643c authored by myhloli's avatar myhloli
Browse files

refactor: streamline bbox processing and enhance category tying logic in magic_model_utils.py

parent ee6d557f
Hide whitespace changes
Inline Side-by-side
Showing with 173 additions and 186 deletions
mineru/backend/pipeline/pipeline_magic_model.py
View file @ 1906643c
from mineru.utils.boxbase import bbox_relative_pos, get_minbox_if_overlap_by_ratio
from mineru.utils.boxbase import bbox_relative_pos, calculate_iou, bbox_distance, is_in, get_minbox_if_overlap_by_ratio
from mineru.utils.enum_class import CategoryId, ContentType
import mineru.utils.magic_model_utils as magic_model_utils
from mineru.utils.magic_model_utils import tie_up_category_by_distance_v3, reduct_overlap
class MagicModel:
......@@ -90,9 +90,18 @@ class MagicModel:
layout_dets.remove(need_remove)
def __fix_by_remove_low_confidence(self):
magic_model_utils.remove_low_confidence(self.__page_model_info['layout_dets'])
need_remove_list = []
layout_dets = self.__page_model_info['layout_dets']
for layout_det in layout_dets:
if layout_det['score'] <= 0.05:
need_remove_list.append(layout_det)
else:
continue
for need_remove in need_remove_list:
layout_dets.remove(need_remove)
def __fix_by_remove_high_iou_and_low_confidence(self):
need_remove_list = []
layout_dets = list(filter(
lambda x: x['category_id'] in [
CategoryId.Title,
......@@ -107,7 +116,20 @@ class MagicModel:
], self.__page_model_info['layout_dets']
)
)
magic_model_utils.remove_high_iou_low_confidence(layout_dets)
for i in range(len(layout_dets)):
for j in range(i + 1, len(layout_dets)):
layout_det1 = layout_dets[i]
layout_det2 = layout_dets[j]
if calculate_iou(layout_det1['bbox'], layout_det2['bbox']) > 0.9:
layout_det_need_remove = layout_det1 if layout_det1['score'] < layout_det2['score'] else layout_det2
if layout_det_need_remove not in need_remove_list:
need_remove_list.append(layout_det_need_remove)
for need_remove in need_remove_list:
self.__page_model_info['layout_dets'].remove(need_remove)
def __fix_footnote(self):
footnotes = []
......@@ -141,7 +163,7 @@ class MagicModel:
if pos_flag_count > 1:
continue
dis_figure_footnote[i] = min(
magic_model_utils.bbox_distance_with_relative_check(figures[j]['bbox'], footnotes[i]['bbox']),
self._bbox_distance(figures[j]['bbox'], footnotes[i]['bbox']),
dis_figure_footnote.get(i, float('inf')),
)
for i in range(len(footnotes)):
......@@ -160,7 +182,7 @@ class MagicModel:
continue
dis_table_footnote[i] = min(
magic_model_utils.bbox_distance_with_relative_check(tables[j]['bbox'], footnotes[i]['bbox']),
self._bbox_distance(tables[j]['bbox'], footnotes[i]['bbox']),
dis_table_footnote.get(i, float('inf')),
)
for i in range(len(footnotes)):
......@@ -169,18 +191,56 @@ class MagicModel:
if dis_table_footnote.get(i, float('inf')) > dis_figure_footnote[i]:
footnotes[i]['category_id'] = CategoryId.ImageFootnote
def __tie_up_category_by_distance_v3(
self,
subject_category_id: int,
object_category_id: int,
):
return magic_model_utils.tie_up_category_by_distance_v3(
self.__page_model_info,
subject_category_id,
object_category_id,
extract_bbox_func=lambda x: x['bbox'],
extract_score_func=lambda x: x['score'],
create_item_func=lambda x: {'bbox': x['bbox'], 'score': x['score']}
def _bbox_distance(self, bbox1, bbox2):
left, right, bottom, top = bbox_relative_pos(bbox1, bbox2)
flags = [left, right, bottom, top]
count = sum([1 if v else 0 for v in flags])
if count > 1:
return float('inf')
if left or right:
l1 = bbox1[3] - bbox1[1]
l2 = bbox2[3] - bbox2[1]
else:
l1 = bbox1[2] - bbox1[0]
l2 = bbox2[2] - bbox2[0]
if l2 > l1 and (l2 - l1) / l1 > 0.3:
return float('inf')
return bbox_distance(bbox1, bbox2)
def __tie_up_category_by_distance_v3(self, subject_category_id, object_category_id):
# 定义获取主体和客体对象的函数
def get_subjects():
return reduct_overlap(
list(
map(
lambda x: {'bbox': x['bbox'], 'score': x['score']},
filter(
lambda x: x['category_id'] == subject_category_id,
self.__page_model_info['layout_dets'],
),
)
)
)
def get_objects():
return reduct_overlap(
list(
map(
lambda x: {'bbox': x['bbox'], 'score': x['score']},
filter(
lambda x: x['category_id'] == object_category_id,
self.__page_model_info['layout_dets'],
),
)
)
)
# 调用通用方法
return tie_up_category_by_distance_v3(
get_subjects,
get_objects
)
def get_imgs(self):
......
mineru/backend/vlm/vlm_magic_model.py
View file @ 1906643c
......@@ -6,7 +6,8 @@ from loguru import logger
from mineru.utils.enum_class import ContentType, BlockType, SplitFlag
from mineru.backend.vlm.vlm_middle_json_mkcontent import merge_para_with_text
from mineru.utils.format_utils import convert_otsl_to_html
import mineru.utils.magic_model_utils as magic_model_utils
from mineru.utils.magic_model_utils import reduct_overlap, tie_up_category_by_distance_v3
class MagicModel:
def __init__(self, token: str, width, height):
......@@ -250,18 +251,38 @@ def latex_fix(latex):
return latex
def __tie_up_category_by_distance_v3(
blocks: list,
subject_block_type: str,
object_block_type: str,
):
return magic_model_utils.tie_up_category_by_distance_v3(
blocks,
lambda x: x["type"] == subject_block_type,
lambda x: x["type"] == object_block_type,
extract_bbox_func=lambda x: x["bbox"],
extract_score_func=lambda x: x.get("score", 1.0),
create_item_func=lambda x: {"bbox": x["bbox"], "lines": x["lines"], "index": x["index"]}
def __tie_up_category_by_distance_v3(blocks, subject_block_type, object_block_type):
# 定义获取主体和客体对象的函数
def get_subjects():
return reduct_overlap(
list(
map(
lambda x: {"bbox": x["bbox"], "lines": x["lines"], "index": x["index"]},
filter(
lambda x: x["type"] == subject_block_type,
blocks,
),
)
)
)
def get_objects():
return reduct_overlap(
list(
map(
lambda x: {"bbox": x["bbox"], "lines": x["lines"], "index": x["index"]},
filter(
lambda x: x["type"] == object_block_type,
blocks,
),
)
)
)
# 调用通用方法
return tie_up_category_by_distance_v3(
get_subjects,
get_objects
)
......
mineru/utils/magic_model_utils.py
View file @ 1906643c
"""
布局处理的公共工具类
包含两个MagicModel类中重复使用的方法和逻辑
"""
from typing import List, Dict, Any, Union
from mineru.utils.boxbase import bbox_relative_pos, calculate_iou, bbox_distance, is_in
from typing import List, Dict, Any, Callable
from mineru.utils.boxbase import bbox_distance, is_in
def reduct_overlap(bboxes: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
......@@ -27,101 +26,48 @@ def reduct_overlap(bboxes: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
return [bboxes[i] for i in range(N) if keep[i]]
def bbox_distance_with_relative_check(bbox1: List[int], bbox2: List[int]) -> float:
"""
计算两个bbox之间的距离，考虑相对位置约束
Args:
bbox1: 第一个bbox [x1, y1, x2, y2]
bbox2: 第二个bbox [x1, y1, x2, y2]
Returns:
距离值，如果不满足条件返回无穷大
"""
left, right, bottom, top = bbox_relative_pos(bbox1, bbox2)
flags = [left, right, bottom, top]
count = sum([1 if v else 0 for v in flags])
if count > 1:
return float('inf')
if left or right:
l1 = bbox1[3] - bbox1[1]
l2 = bbox2[3] - bbox2[1]
else:
l1 = bbox1[2] - bbox1[0]
l2 = bbox2[2] - bbox2[0]
if l2 > l1 and (l2 - l1) / l1 > 0.3:
return float('inf')
return bbox_distance(bbox1, bbox2)
def tie_up_category_by_distance_v3(
data_source: Union[List[Dict], Dict],
subject_category_filter,
object_category_filter,
extract_bbox_func=None,
extract_score_func=None,
create_item_func=None
) -> List[Dict[str, Any]]:
get_subjects_func: Callable,
get_objects_func: Callable,
extract_subject_func: Callable = None,
extract_object_func: Callable = None
):
"""
基于距离关联不同类型的区块/元素
通用的类别关联方法，用于将主体对象与客体对象进行关联
Args:
data_source: 数据源，可以是列表或包含layout_dets的字典
subject_category_filter: 主体类别过滤函数或值
object_category_filter: 对象类别过滤函数或值
extract_bbox_func: 提取bbox的函数，默认使用'bbox'键
extract_score_func: 提取score的函数，默认使用'score'键
create_item_func: 创建返回项的函数
参数:
get_subjects_func: 函数，提取主体对象
get_objects_func: 函数，提取客体对象
extract_subject_func: 函数，自定义提取主体属性（默认使用bbox和其他属性）
extract_object_func: 函数，自定义提取客体属性（默认使用bbox和其他属性）
Returns:
关联结果列表
返回:
关联后的对象列表
"""
# 默认函数
if extract_bbox_func is None:
extract_bbox_func = lambda x: x['bbox']
if extract_score_func is None:
extract_score_func = lambda x: x['score']
if create_item_func is None:
create_item_func = lambda x: {'bbox': extract_bbox_func(x), 'score': extract_score_func(x)}
# 处理数据源
if isinstance(data_source, dict) and 'layout_dets' in data_source:
items = data_source['layout_dets']
else:
items = data_source
subjects = get_subjects_func()
objects = get_objects_func()
# 过滤主体和对象
if callable(subject_category_filter):
subjects = list(filter(subject_category_filter, items))
else:
subjects = list(filter(lambda x: x.get('category_id') == subject_category_filter or x.get('type') == subject_category_filter, items))
if callable(object_category_filter):
objects = list(filter(object_category_filter, items))
else:
objects = list(filter(lambda x: x.get('category_id') == object_category_filter or x.get('type') == object_category_filter, items))
# 转换为标准格式并去重
subjects = reduct_overlap([create_item_func(x) for x in subjects])
objects = reduct_overlap([create_item_func(x) for x in objects])
# 如果没有提供自定义提取函数，使用默认函数
if extract_subject_func is None:
extract_subject_func = lambda x: x
if extract_object_func is None:
extract_object_func = lambda x: x
ret = []
N, M = len(subjects), len(objects)
subjects.sort(key=lambda x: extract_bbox_func(x)[0] ** 2 + extract_bbox_func(x)[1] ** 2)
objects.sort(key=lambda x: extract_bbox_func(x)[0] ** 2 + extract_bbox_func(x)[1] ** 2)
subjects.sort(key=lambda x: x["bbox"][0] ** 2 + x["bbox"][1] ** 2)
objects.sort(key=lambda x: x["bbox"][0] ** 2 + x["bbox"][1] ** 2)
OBJ_IDX_OFFSET = 10000
SUB_BIT_KIND, OBJ_BIT_KIND = 0, 1
all_boxes_with_idx = [(i, SUB_BIT_KIND, extract_bbox_func(sub)[0], extract_bbox_func(sub)[1]) for i, sub in enumerate(subjects)] + \
[(i + OBJ_IDX_OFFSET, OBJ_BIT_KIND, extract_bbox_func(obj)[0], extract_bbox_func(obj)[1]) for i, obj in enumerate(objects)]
all_boxes_with_idx = [(i, SUB_BIT_KIND, sub["bbox"][0], sub["bbox"][1]) for i, sub in enumerate(subjects)] + [
(i + OBJ_IDX_OFFSET, OBJ_BIT_KIND, obj["bbox"][0], obj["bbox"][1]) for i, obj in enumerate(objects)
]
seen_idx = set()
seen_sub_idx = set()
seen_sub_idx_len = len(seen_sub_idx)
while N > seen_sub_idx_len:
while N > len(seen_sub_idx):
candidates = []
for idx, kind, x0, y0 in all_boxes_with_idx:
if idx in seen_idx:
......@@ -136,10 +82,10 @@ def tie_up_category_by_distance_v3(
candidates.sort(key=lambda x: (x[2] - left_x) ** 2 + (x[3] - top_y) ** 2)
fst_idx, fst_kind, left_x, top_y = candidates[0]
fst_bbox = extract_bbox_func(subjects[fst_idx]) if fst_kind == SUB_BIT_KIND else extract_bbox_func(objects[fst_idx - OBJ_IDX_OFFSET])
fst_bbox = subjects[fst_idx]['bbox'] if fst_kind == SUB_BIT_KIND else objects[fst_idx - OBJ_IDX_OFFSET]['bbox']
candidates.sort(
key=lambda x: bbox_distance(fst_bbox, extract_bbox_func(subjects[x[0]])) if x[1] == SUB_BIT_KIND else bbox_distance(
fst_bbox, extract_bbox_func(objects[x[0] - OBJ_IDX_OFFSET])))
key=lambda x: bbox_distance(fst_bbox, subjects[x[0]]['bbox']) if x[1] == SUB_BIT_KIND else bbox_distance(
fst_bbox, objects[x[0] - OBJ_IDX_OFFSET]['bbox']))
nxt = None
for i in range(1, len(candidates)):
......@@ -154,12 +100,12 @@ def tie_up_category_by_distance_v3(
else:
sub_idx, obj_idx = nxt[0], fst_idx - OBJ_IDX_OFFSET
pair_dis = bbox_distance(extract_bbox_func(subjects[sub_idx]), extract_bbox_func(objects[obj_idx]))
nearest_dis = float('inf')
pair_dis = bbox_distance(subjects[sub_idx]["bbox"], objects[obj_idx]["bbox"])
nearest_dis = float("inf")
for i in range(N):
# 取消原先算法中 1对1 匹配的偏置
# if i in seen_idx or i == sub_idx:continue
nearest_dis = min(nearest_dis, bbox_distance(extract_bbox_func(subjects[i]), extract_bbox_func(objects[obj_idx])))
nearest_dis = min(nearest_dis, bbox_distance(subjects[i]["bbox"], objects[obj_idx]["bbox"]))
if pair_dis >= 3 * nearest_dis:
seen_idx.add(sub_idx)
......@@ -169,21 +115,22 @@ def tie_up_category_by_distance_v3(
seen_idx.add(obj_idx + OBJ_IDX_OFFSET)
seen_sub_idx.add(sub_idx)
ret.append({
'sub_bbox': subjects[sub_idx],
'obj_bboxes': [objects[obj_idx]],
'sub_idx': sub_idx,
})
ret.append(
{
"sub_bbox": extract_subject_func(subjects[sub_idx]),
"obj_bboxes": [extract_object_func(objects[obj_idx])],
"sub_idx": sub_idx,
}
)
# 处理剩余的对象
for i in range(len(objects)):
j = i + OBJ_IDX_OFFSET
if j in seen_idx:
continue
seen_idx.add(j)
nearest_dis, nearest_sub_idx = float('inf'), -1
nearest_dis, nearest_sub_idx = float("inf"), -1
for k in range(len(subjects)):
dis = bbox_distance(extract_bbox_func(objects[i]), extract_bbox_func(subjects[k]))
dis = bbox_distance(objects[i]["bbox"], subjects[k]["bbox"])
if dis < nearest_dis:
nearest_dis = dis
nearest_sub_idx = k
......@@ -193,70 +140,29 @@ def tie_up_category_by_distance_v3(
continue
if k in seen_sub_idx:
for kk in range(len(ret)):
if ret[kk]['sub_idx'] == k:
ret[kk]['obj_bboxes'].append(objects[i])
if ret[kk]["sub_idx"] == k:
ret[kk]["obj_bboxes"].append(extract_object_func(objects[i]))
break
else:
ret.append({
'sub_bbox': subjects[k],
'obj_bboxes': [objects[i]],
'sub_idx': k,
})
ret.append(
{
"sub_bbox": extract_subject_func(subjects[k]),
"obj_bboxes": [extract_object_func(objects[i])],
"sub_idx": k,
}
)
seen_sub_idx.add(k)
seen_idx.add(k)
# 处理剩余的主体
for i in range(len(subjects)):
if i in seen_sub_idx:
continue
ret.append({
'sub_bbox': subjects[i],
'obj_bboxes': [],
'sub_idx': i,
})
return ret
def remove_high_iou_low_confidence(layout_dets: List[Dict], iou_threshold: float = 0.9):
"""
删除高IOU且置信度较低的检测结果
Args:
layout_dets: 布局检测结果列表
iou_threshold: IOU阈值
"""
need_remove_list = []
for i in range(len(layout_dets)):
for j in range(i + 1, len(layout_dets)):
layout_det1 = layout_dets[i]
layout_det2 = layout_dets[j]
if calculate_iou(layout_det1['bbox'], layout_det2['bbox']) > iou_threshold:
layout_det_need_remove = layout_det1 if layout_det1['score'] < layout_det2['score'] else layout_det2
if layout_det_need_remove not in need_remove_list:
need_remove_list.append(layout_det_need_remove)
for need_remove in need_remove_list:
if need_remove in layout_dets:
layout_dets.remove(need_remove)
def remove_low_confidence(layout_dets: List[Dict], confidence_threshold: float = 0.05):
"""
删除置信度特别低的检测结果
Args:
layout_dets: 布局检测结果列表
confidence_threshold: 置信度阈值
"""
need_remove_list = []
for layout_det in layout_dets:
if layout_det['score'] <= confidence_threshold:
need_remove_list.append(layout_det)
for need_remove in need_remove_list:
if need_remove in layout_dets:
layout_dets.remove(need_remove)
ret.append(
{
"sub_bbox": extract_subject_func(subjects[i]),
"obj_bboxes": [],
"sub_idx": i,
}
)
return ret
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment