magic_model.py 24.2 KB
Newer Older
liukaiwen's avatar
liukaiwen committed
1
2
import json
import math
kernel.h@qq.com's avatar
kernel.h@qq.com committed
3

liukaiwen's avatar
liukaiwen committed
4
5
from magic_pdf.libs.commons import fitz
from loguru import logger
kernel.h@qq.com's avatar
kernel.h@qq.com committed
6

liukaiwen's avatar
liukaiwen committed
7
8
9
10
11
12
from magic_pdf.libs.commons import join_path
from magic_pdf.libs.coordinate_transform import get_scale_ratio
from magic_pdf.libs.ocr_content_type import ContentType
from magic_pdf.rw.AbsReaderWriter import AbsReaderWriter
from magic_pdf.rw.DiskReaderWriter import DiskReaderWriter
from magic_pdf.libs.math import float_gt
13
14
15
16
17
from magic_pdf.libs.boxbase import (
    _is_in,
    bbox_relative_pos,
    bbox_distance,
    _is_part_overlap,
18
    calculate_overlap_area_in_bbox1_area_ratio, calculate_iou,
19
)
liukaiwen's avatar
liukaiwen committed
20
from magic_pdf.libs.ModelBlockTypeEnum import ModelBlockTypeEnum
liukaiwen's avatar
liukaiwen committed
21

22
CAPATION_OVERLAP_AREA_RATIO = 0.6
liukaiwen's avatar
liukaiwen committed
23

许瑞's avatar
许瑞 committed
24

liukaiwen's avatar
liukaiwen committed
25
class MagicModel:
kernel.h@qq.com's avatar
kernel.h@qq.com committed
26
27
    """
    每个函数没有得到元素的时候返回空list
liukaiwen's avatar
liukaiwen committed
28

kernel.h@qq.com's avatar
kernel.h@qq.com committed
29
    """
liukaiwen's avatar
liukaiwen committed
30
31
32

    def __fix_axis(self):
        for model_page_info in self.__model_list:
33
            need_remove_list = []
liukaiwen's avatar
liukaiwen committed
34
35
36
37
38
39
40
41
42
43
44
45
46
47
            page_no = model_page_info["page_info"]["page_no"]
            horizontal_scale_ratio, vertical_scale_ratio = get_scale_ratio(
                model_page_info, self.__docs[page_no]
            )
            layout_dets = model_page_info["layout_dets"]
            for layout_det in layout_dets:
                x0, y0, _, _, x1, y1, _, _ = layout_det["poly"]
                bbox = [
                    int(x0 / horizontal_scale_ratio),
                    int(y0 / vertical_scale_ratio),
                    int(x1 / horizontal_scale_ratio),
                    int(y1 / vertical_scale_ratio),
                ]
                layout_det["bbox"] = bbox
48
49
                # 删除高度或者宽度小于等于0的spans
                if bbox[2] - bbox[0] <= 0 or bbox[3] - bbox[1] <= 0:
liukaiwen's avatar
liukaiwen committed
50
51
52
53
                    need_remove_list.append(layout_det)
            for need_remove in need_remove_list:
                layout_dets.remove(need_remove)

54
    def __fix_by_remove_low_confidence(self):
55
56
57
58
        for model_page_info in self.__model_list:
            need_remove_list = []
            layout_dets = model_page_info["layout_dets"]
            for layout_det in layout_dets:
赵小蒙's avatar
赵小蒙 committed
59
                if layout_det["score"] <= 0.05:
60
61
62
63
64
                    need_remove_list.append(layout_det)
                else:
                    continue
            for need_remove in need_remove_list:
                layout_dets.remove(need_remove)
liukaiwen's avatar
liukaiwen committed
65

66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
    def __fix_by_remove_high_iou_and_low_confidence(self):
        for model_page_info in self.__model_list:
            need_remove_list = []
            layout_dets = model_page_info["layout_dets"]
            for layout_det1 in layout_dets:
                for layout_det2 in layout_dets:
                    if layout_det1 == layout_det2:
                        continue
                    if layout_det1["category_id"] in [0,1,2,3,4,5,6,7,8,9] and layout_det2["category_id"] in [0,1,2,3,4,5,6,7,8,9]:
                        if calculate_iou(layout_det1['bbox'], layout_det2['bbox']) > 0.9:
                            if layout_det1['score'] < layout_det2['score']:
                                layout_det_need_remove = layout_det1
                            else:
                                layout_det_need_remove = layout_det2

                            if layout_det_need_remove not in need_remove_list:
                                need_remove_list.append(layout_det_need_remove)
                        else:
                            continue
                    else:
                        continue
            for need_remove in need_remove_list:
                layout_dets.remove(need_remove)

liukaiwen's avatar
liukaiwen committed
90
    def __init__(self, model_list: list, docs: fitz.Document):
kernel.h@qq.com's avatar
kernel.h@qq.com committed
91
        self.__model_list = model_list
liukaiwen's avatar
liukaiwen committed
92
        self.__docs = docs
赵小蒙's avatar
赵小蒙 committed
93
        '''为所有模型数据添加bbox信息(缩放,poly->bbox)'''
kernel.h@qq.com's avatar
kernel.h@qq.com committed
94
        self.__fix_axis()
赵小蒙's avatar
赵小蒙 committed
95
        '''删除置信度特别低的模型数据(<0.05),提高质量'''
96
        self.__fix_by_remove_low_confidence()
赵小蒙's avatar
赵小蒙 committed
97
        '''删除高iou(>0.9)数据中置信度较低的那个'''
98
        self.__fix_by_remove_high_iou_and_low_confidence()
liukaiwen's avatar
liukaiwen committed
99
100
101
102
103
104
105
106

    def __reduct_overlap(self, bboxes):
        N = len(bboxes)
        keep = [True] * N
        for i in range(N):
            for j in range(N):
                if i == j:
                    continue
许瑞's avatar
许瑞 committed
107
                if _is_in(bboxes[i]["bbox"], bboxes[j]["bbox"]):
liukaiwen's avatar
liukaiwen committed
108
109
110
111
112
                    keep[i] = False

        return [bboxes[i] for i in range(N) if keep[i]]

    def __tie_up_category_by_distance(
113
        self, page_no, subject_category_id, object_category_id
liukaiwen's avatar
liukaiwen committed
114
115
116
117
118
    ):
        """
        假定每个 subject 最多有一个 object (可以有多个相邻的 object 合并为单个 object),每个 object 只能属于一个 subject
        """
        ret = []
119
        MAX_DIS_OF_POINT = 10**9 + 7
liukaiwen's avatar
liukaiwen committed
120

许瑞's avatar
许瑞 committed
121
122
123
124
125
126
127
128
129
130
        def expand_bbox(bbox1, bbox2):
            x0 = min(bbox1[0], bbox2[0])
            y0 = min(bbox1[1], bbox2[1])
            x1 = max(bbox1[2], bbox2[2])
            y1 = max(bbox1[3], bbox2[3])
            return [x0, y0, x1, y1]

        def get_bbox_area(bbox):
            return abs(bbox[2] - bbox[0]) * abs(bbox[3] - bbox[1])

许瑞's avatar
许瑞 committed
131
132
133
134
        # subject 和 object 的 bbox 会合并成一个大的 bbox (named: merged bbox)。 筛选出所有和 merged bbox 有 overlap 且 overlap 面积大于 object 的面积的 subjects。
        # 再求出筛选出的 subjects 和 object 的最短距离!
        def may_find_other_nearest_bbox(subject_idx, object_idx):
            ret = float("inf")
135

许瑞's avatar
许瑞 committed
136
137
138
139
140
141
142
143
144
145
146
147
            x0 = min(
                all_bboxes[subject_idx]["bbox"][0], all_bboxes[object_idx]["bbox"][0]
            )
            y0 = min(
                all_bboxes[subject_idx]["bbox"][1], all_bboxes[object_idx]["bbox"][1]
            )
            x1 = max(
                all_bboxes[subject_idx]["bbox"][2], all_bboxes[object_idx]["bbox"][2]
            )
            y1 = max(
                all_bboxes[subject_idx]["bbox"][3], all_bboxes[object_idx]["bbox"][3]
            )
许瑞's avatar
许瑞 committed
148

许瑞's avatar
许瑞 committed
149
150
151
152
153
            object_area = abs(
                all_bboxes[object_idx]["bbox"][2] - all_bboxes[object_idx]["bbox"][0]
            ) * abs(
                all_bboxes[object_idx]["bbox"][3] - all_bboxes[object_idx]["bbox"][1]
            )
许瑞's avatar
许瑞 committed
154
155

            for i in range(len(all_bboxes)):
许瑞's avatar
许瑞 committed
156
157
158
159
                if (
                    i == subject_idx
                    or all_bboxes[i]["category_id"] != subject_category_id
                ):
许瑞's avatar
许瑞 committed
160
                    continue
许瑞's avatar
许瑞 committed
161
162
163
                if _is_part_overlap([x0, y0, x1, y1], all_bboxes[i]["bbox"]) or _is_in(
                    all_bboxes[i]["bbox"], [x0, y0, x1, y1]
                ):
164

许瑞's avatar
许瑞 committed
165
166
167
                    i_area = abs(
                        all_bboxes[i]["bbox"][2] - all_bboxes[i]["bbox"][0]
                    ) * abs(all_bboxes[i]["bbox"][3] - all_bboxes[i]["bbox"][1])
许瑞's avatar
许瑞 committed
168
                    if i_area >= object_area:
许瑞's avatar
许瑞 committed
169
                        ret = min(float("inf"), dis[i][object_idx])
170

许瑞's avatar
许瑞 committed
171
172
            return ret

liukaiwen's avatar
liukaiwen committed
173
174
175
        subjects = self.__reduct_overlap(
            list(
                map(
许瑞's avatar
许瑞 committed
176
                    lambda x: {"bbox": x["bbox"], "score": x["score"]},
liukaiwen's avatar
liukaiwen committed
177
178
179
180
181
182
183
184
185
186
187
                    filter(
                        lambda x: x["category_id"] == subject_category_id,
                        self.__model_list[page_no]["layout_dets"],
                    ),
                )
            )
        )

        objects = self.__reduct_overlap(
            list(
                map(
许瑞's avatar
许瑞 committed
188
                    lambda x: {"bbox": x["bbox"], "score": x["score"]},
liukaiwen's avatar
liukaiwen committed
189
190
191
192
193
194
195
196
197
                    filter(
                        lambda x: x["category_id"] == object_category_id,
                        self.__model_list[page_no]["layout_dets"],
                    ),
                )
            )
        )
        subject_object_relation_map = {}

许瑞's avatar
许瑞 committed
198
199
200
        subjects.sort(
            key=lambda x: x["bbox"][0] ** 2 + x["bbox"][1] ** 2
        )  # get the distance !
liukaiwen's avatar
liukaiwen committed
201
202
203
204

        all_bboxes = []

        for v in subjects:
许瑞's avatar
许瑞 committed
205
206
207
208
209
210
211
            all_bboxes.append(
                {
                    "category_id": subject_category_id,
                    "bbox": v["bbox"],
                    "score": v["score"],
                }
            )
liukaiwen's avatar
liukaiwen committed
212
213

        for v in objects:
许瑞's avatar
许瑞 committed
214
215
216
217
218
219
220
            all_bboxes.append(
                {
                    "category_id": object_category_id,
                    "bbox": v["bbox"],
                    "score": v["score"],
                }
            )
liukaiwen's avatar
liukaiwen committed
221
222
223
224
225
226
227

        N = len(all_bboxes)
        dis = [[MAX_DIS_OF_POINT] * N for _ in range(N)]

        for i in range(N):
            for j in range(i):
                if (
228
229
                    all_bboxes[i]["category_id"] == subject_category_id
                    and all_bboxes[j]["category_id"] == subject_category_id
liukaiwen's avatar
liukaiwen committed
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
                ):
                    continue

                dis[i][j] = bbox_distance(all_bboxes[i]["bbox"], all_bboxes[j]["bbox"])
                dis[j][i] = dis[i][j]

        used = set()
        for i in range(N):
            # 求第 i 个 subject 所关联的 object
            if all_bboxes[i]["category_id"] != subject_category_id:
                continue
            seen = set()
            candidates = []
            arr = []
            for j in range(N):

                pos_flag_count = sum(
                    list(
                        map(
                            lambda x: 1 if x else 0,
                            bbox_relative_pos(
                                all_bboxes[i]["bbox"], all_bboxes[j]["bbox"]
                            ),
                        )
                    )
                )
                if pos_flag_count > 1:
                    continue
                if (
259
260
261
                    all_bboxes[j]["category_id"] != object_category_id
                    or j in used
                    or dis[i][j] == MAX_DIS_OF_POINT
liukaiwen's avatar
liukaiwen committed
262
263
                ):
                    continue
许瑞's avatar
许瑞 committed
264
265
266
267
268
269
270
                left, right, _, _ = bbox_relative_pos(all_bboxes[i]["bbox"], all_bboxes[j]["bbox"]) # 由  pos_flag_count 相关逻辑保证本段逻辑准确性
                if left or right:
                    one_way_dis = all_bboxes[i]["bbox"][2] - all_bboxes[i]["bbox"][0]
                else:
                    one_way_dis = all_bboxes[i]["bbox"][3] - all_bboxes[i]["bbox"][1]
                if dis[i][j] > one_way_dis:
                    continue
liukaiwen's avatar
liukaiwen committed
271
272
273
274
                arr.append((dis[i][j], j))

            arr.sort(key=lambda x: x[0])
            if len(arr) > 0:
许瑞's avatar
许瑞 committed
275
                # bug: 离该subject 最近的 object 可能跨越了其它的 subject 。比如 [this subect] [some sbuject] [the nearest objec of subject]
许瑞's avatar
许瑞 committed
276
                if may_find_other_nearest_bbox(i, arr[0][1]) >= arr[0][0]:
许瑞's avatar
许瑞 committed
277

许瑞's avatar
许瑞 committed
278
279
                    candidates.append(arr[0][1])
                    seen.add(arr[0][1])
liukaiwen's avatar
liukaiwen committed
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299

            # 已经获取初始种子
            for j in set(candidates):
                tmp = []
                for k in range(i + 1, N):
                    pos_flag_count = sum(
                        list(
                            map(
                                lambda x: 1 if x else 0,
                                bbox_relative_pos(
                                    all_bboxes[j]["bbox"], all_bboxes[k]["bbox"]
                                ),
                            )
                        )
                    )

                    if pos_flag_count > 1:
                        continue

                    if (
300
301
302
303
                        all_bboxes[k]["category_id"] != object_category_id
                        or k in used
                        or k in seen
                        or dis[j][k] == MAX_DIS_OF_POINT
许瑞's avatar
许瑞 committed
304
                        or dis[j][k] > dis[i][j]
liukaiwen's avatar
liukaiwen committed
305
306
                    ):
                        continue
许瑞's avatar
许瑞 committed
307

liukaiwen's avatar
liukaiwen committed
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
                    is_nearest = True
                    for l in range(i + 1, N):
                        if l in (j, k) or l in used or l in seen:
                            continue

                        if not float_gt(dis[l][k], dis[j][k]):
                            is_nearest = False
                            break

                    if is_nearest:
                        tmp.append(k)
                        seen.add(k)

                candidates = tmp
                if len(candidates) == 0:
                    break

            # 已经获取到某个 figure 下所有的最靠近的 captions,以及最靠近这些 captions 的 captions 。
            # 先扩一下 bbox,
            x0s = [all_bboxes[idx]["bbox"][0] for idx in seen] + [
                all_bboxes[i]["bbox"][0]
            ]
            y0s = [all_bboxes[idx]["bbox"][1] for idx in seen] + [
                all_bboxes[i]["bbox"][1]
            ]
            x1s = [all_bboxes[idx]["bbox"][2] for idx in seen] + [
                all_bboxes[i]["bbox"][2]
            ]
            y1s = [all_bboxes[idx]["bbox"][3] for idx in seen] + [
                all_bboxes[i]["bbox"][3]
            ]

            ox0, oy0, ox1, oy1 = min(x0s), min(y0s), max(x1s), max(y1s)
            ix0, iy0, ix1, iy1 = all_bboxes[i]["bbox"]

            # 分成了 4 个截取空间,需要计算落在每个截取空间下 objects 合并后占据的矩形面积
            caption_poses = [
                [ox0, oy0, ix0, oy1],
                [ox0, oy0, ox1, iy0],
                [ox0, iy1, ox1, oy1],
                [ix1, oy0, ox1, oy1],
            ]

            caption_areas = []
            for bbox in caption_poses:
                embed_arr = []
                for idx in seen:
许瑞's avatar
许瑞 committed
355
356
357
358
359
360
                    if (
                        calculate_overlap_area_in_bbox1_area_ratio(
                            all_bboxes[idx]["bbox"], bbox
                        )
                        > CAPATION_OVERLAP_AREA_RATIO
                    ):
liukaiwen's avatar
liukaiwen committed
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
                        embed_arr.append(idx)

                if len(embed_arr) > 0:
                    embed_x0 = min([all_bboxes[idx]["bbox"][0] for idx in embed_arr])
                    embed_y0 = min([all_bboxes[idx]["bbox"][1] for idx in embed_arr])
                    embed_x1 = max([all_bboxes[idx]["bbox"][2] for idx in embed_arr])
                    embed_y1 = max([all_bboxes[idx]["bbox"][3] for idx in embed_arr])
                    caption_areas.append(
                        int(abs(embed_x1 - embed_x0) * abs(embed_y1 - embed_y0))
                    )
                else:
                    caption_areas.append(0)

            subject_object_relation_map[i] = []
            if max(caption_areas) > 0:
                max_area_idx = caption_areas.index(max(caption_areas))
                caption_bbox = caption_poses[max_area_idx]

                for j in seen:
许瑞's avatar
许瑞 committed
380
381
382
383
384
385
                    if (
                        calculate_overlap_area_in_bbox1_area_ratio(
                            all_bboxes[j]["bbox"], caption_bbox
                        )
                        > CAPATION_OVERLAP_AREA_RATIO
                    ):
liukaiwen's avatar
liukaiwen committed
386
387
388
389
390
391
392
                        used.add(j)
                        subject_object_relation_map[i].append(j)

        for i in sorted(subject_object_relation_map.keys()):
            result = {
                "subject_body": all_bboxes[i]["bbox"],
                "all": all_bboxes[i]["bbox"],
许瑞's avatar
许瑞 committed
393
                "score": all_bboxes[i]["score"],
liukaiwen's avatar
liukaiwen committed
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
            }

            if len(subject_object_relation_map[i]) > 0:
                x0 = min(
                    [all_bboxes[j]["bbox"][0] for j in subject_object_relation_map[i]]
                )
                y0 = min(
                    [all_bboxes[j]["bbox"][1] for j in subject_object_relation_map[i]]
                )
                x1 = max(
                    [all_bboxes[j]["bbox"][2] for j in subject_object_relation_map[i]]
                )
                y1 = max(
                    [all_bboxes[j]["bbox"][3] for j in subject_object_relation_map[i]]
                )
                result["object_body"] = [x0, y0, x1, y1]
                result["all"] = [
                    min(x0, all_bboxes[i]["bbox"][0]),
                    min(y0, all_bboxes[i]["bbox"][1]),
                    max(x1, all_bboxes[i]["bbox"][2]),
                    max(y1, all_bboxes[i]["bbox"][3]),
                ]
            ret.append(result)

        total_subject_object_dis = 0
        # 计算已经配对的 distance 距离
        for i in subject_object_relation_map.keys():
            for j in subject_object_relation_map[i]:
                total_subject_object_dis += bbox_distance(
                    all_bboxes[i]["bbox"], all_bboxes[j]["bbox"]
                )

        # 计算未匹配的 subject 和 object 的距离(非精确版)
        with_caption_subject = set(
            [
                key
                for key in subject_object_relation_map.keys()
                if len(subject_object_relation_map[i]) > 0
            ]
        )
        for i in range(N):
            if all_bboxes[i]["category_id"] != object_category_id or i in used:
                continue
            candidates = []
            for j in range(N):
                if (
440
441
                    all_bboxes[j]["category_id"] != subject_category_id
                    or j in with_caption_subject
liukaiwen's avatar
liukaiwen committed
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
                ):
                    continue
                candidates.append((dis[i][j], j))
            if len(candidates) > 0:
                candidates.sort(key=lambda x: x[0])
                total_subject_object_dis += candidates[0][1]
                with_caption_subject.add(j)
        return ret, total_subject_object_dis

    def get_imgs(self, page_no: int):  # @许瑞
        records, _ = self.__tie_up_category_by_distance(page_no, 3, 4)
        return [
            {
                "bbox": record["all"],
                "img_body_bbox": record["subject_body"],
                "img_caption_bbox": record.get("object_body", None),
许瑞's avatar
许瑞 committed
458
                "score": record["score"],
liukaiwen's avatar
liukaiwen committed
459
460
461
462
463
            }
            for record in records
        ]

    def get_tables(
464
        self, page_no: int
liukaiwen's avatar
liukaiwen committed
465
466
467
468
469
470
471
472
    ) -> list:  # 3个坐标, caption, table主体,table-note
        with_captions, _ = self.__tie_up_category_by_distance(page_no, 5, 6)
        with_footnotes, _ = self.__tie_up_category_by_distance(page_no, 5, 7)
        ret = []
        N, M = len(with_captions), len(with_footnotes)
        assert N == M
        for i in range(N):
            record = {
许瑞's avatar
许瑞 committed
473
                "score": with_captions[i]["score"],
liukaiwen's avatar
liukaiwen committed
474
475
476
477
478
479
480
481
482
483
484
485
486
487
                "table_caption_bbox": with_captions[i].get("object_body", None),
                "table_body_bbox": with_captions[i]["subject_body"],
                "table_footnote_bbox": with_footnotes[i].get("object_body", None),
            }

            x0 = min(with_captions[i]["all"][0], with_footnotes[i]["all"][0])
            y0 = min(with_captions[i]["all"][1], with_footnotes[i]["all"][1])
            x1 = max(with_captions[i]["all"][2], with_footnotes[i]["all"][2])
            y1 = max(with_captions[i]["all"][3], with_footnotes[i]["all"][3])
            record["bbox"] = [x0, y0, x1, y1]
            ret.append(record)
        return ret

    def get_equations(self, page_no: int) -> list:  # 有坐标,也有字
488
489
490
491
492
493
494
495
496
        inline_equations = self.__get_blocks_by_type(
            ModelBlockTypeEnum.EMBEDDING.value, page_no, ["latex"]
        )
        interline_equations = self.__get_blocks_by_type(
            ModelBlockTypeEnum.ISOLATED.value, page_no, ["latex"]
        )
        interline_equations_blocks = self.__get_blocks_by_type(
            ModelBlockTypeEnum.ISOLATE_FORMULA.value, page_no
        )
liukaiwen's avatar
liukaiwen committed
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
        return inline_equations, interline_equations, interline_equations_blocks

    def get_discarded(self, page_no: int) -> list:  # 自研模型,只有坐标
        blocks = self.__get_blocks_by_type(ModelBlockTypeEnum.ABANDON.value, page_no)
        return blocks

    def get_text_blocks(self, page_no: int) -> list:  # 自研模型搞的,只有坐标,没有字
        blocks = self.__get_blocks_by_type(ModelBlockTypeEnum.PLAIN_TEXT.value, page_no)
        return blocks

    def get_title_blocks(self, page_no: int) -> list:  # 自研模型,只有坐标,没字
        blocks = self.__get_blocks_by_type(ModelBlockTypeEnum.TITLE.value, page_no)
        return blocks

    def get_ocr_text(self, page_no: int) -> list:  # paddle 搞的,有字也有坐标
        text_spans = []
        model_page_info = self.__model_list[page_no]
        layout_dets = model_page_info["layout_dets"]
        for layout_det in layout_dets:
            if layout_det["category_id"] == "15":
                span = {
518
                    "bbox": layout_det["bbox"],
liukaiwen's avatar
liukaiwen committed
519
520
521
522
523
524
                    "content": layout_det["text"],
                }
                text_spans.append(span)
        return text_spans

    def get_all_spans(self, page_no: int) -> list:
525
526
527
528
529
530
        def remove_duplicate_spans(spans):
            new_spans = []
            for span in spans:
                if not any(span == existing_span for existing_span in new_spans):
                    new_spans.append(span)
            return new_spans
liukaiwen's avatar
liukaiwen committed
531
532
533
534
535
536
        all_spans = []
        model_page_info = self.__model_list[page_no]
        layout_dets = model_page_info["layout_dets"]
        allow_category_id_list = [3, 5, 13, 14, 15]
        """当成span拼接的"""
        #  3: 'image', # 图片
537
        #  5: 'table',       # 表格
liukaiwen's avatar
liukaiwen committed
538
539
540
541
542
543
        #  13: 'inline_equation',     # 行内公式
        #  14: 'interline_equation',      # 行间公式
        #  15: 'text',      # ocr识别文本
        for layout_det in layout_dets:
            category_id = layout_det["category_id"]
            if category_id in allow_category_id_list:
544
545
546
547
                span = {
                    "bbox": layout_det["bbox"],
                    "score": layout_det["score"]
                }
liukaiwen's avatar
liukaiwen committed
548
549
550
551
552
553
554
555
556
557
558
559
560
561
                if category_id == 3:
                    span["type"] = ContentType.Image
                elif category_id == 5:
                    span["type"] = ContentType.Table
                elif category_id == 13:
                    span["content"] = layout_det["latex"]
                    span["type"] = ContentType.InlineEquation
                elif category_id == 14:
                    span["content"] = layout_det["latex"]
                    span["type"] = ContentType.InterlineEquation
                elif category_id == 15:
                    span["content"] = layout_det["text"]
                    span["type"] = ContentType.Text
                all_spans.append(span)
562
        return remove_duplicate_spans(all_spans)
liukaiwen's avatar
liukaiwen committed
563
564
565
566
567
568
569
570
571

    def get_page_size(self, page_no: int):  # 获取页面宽高
        # 获取当前页的page对象
        page = self.__docs[page_no]
        # 获取当前页的宽高
        page_w = page.rect.width
        page_h = page.rect.height
        return page_w, page_h

572
573
574
    def __get_blocks_by_type(
        self, type: int, page_no: int, extra_col: list[str] = []
    ) -> list:
liukaiwen's avatar
liukaiwen committed
575
576
577
578
579
580
581
582
583
584
585
        blocks = []
        for page_dict in self.__model_list:
            layout_dets = page_dict.get("layout_dets", [])
            page_info = page_dict.get("page_info", {})
            page_number = page_info.get("page_no", -1)
            if page_no != page_number:
                continue
            for item in layout_dets:
                category_id = item.get("category_id", -1)
                bbox = item.get("bbox", None)

liukaiwen's avatar
liukaiwen committed
586
                if category_id == type:
587
588
589
590
                    block = {
                        "bbox": bbox,
                        "score": item.get("score"),
                    }
liukaiwen's avatar
liukaiwen committed
591
592
593
594
595
                    for col in extra_col:
                        block[col] = item.get(col, None)
                    blocks.append(block)
        return blocks

许瑞's avatar
许瑞 committed
596
597
598
    def get_model_list(self, page_no):
        return self.__model_list[page_no]

599
600


liukaiwen's avatar
liukaiwen committed
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
if __name__ == "__main__":
    drw = DiskReaderWriter(r"D:/project/20231108code-clean")
    if 0:
        pdf_file_path = r"linshixuqiu\19983-00.pdf"
        model_file_path = r"linshixuqiu\19983-00_new.json"
        pdf_bytes = drw.read(pdf_file_path, AbsReaderWriter.MODE_BIN)
        model_json_txt = drw.read(model_file_path, AbsReaderWriter.MODE_TXT)
        model_list = json.loads(model_json_txt)
        write_path = r"D:\project\20231108code-clean\linshixuqiu\19983-00"
        img_bucket_path = "imgs"
        img_writer = DiskReaderWriter(join_path(write_path, img_bucket_path))
        pdf_docs = fitz.open("pdf", pdf_bytes)
        magic_model = MagicModel(model_list, pdf_docs)

    if 1:
        model_list = json.loads(
            drw.read("/opt/data/pdf/20240418/j.chroma.2009.03.042.json")
        )
        pdf_bytes = drw.read(
            "/opt/data/pdf/20240418/j.chroma.2009.03.042.pdf", AbsReaderWriter.MODE_BIN
        )
        pdf_docs = fitz.open("pdf", pdf_bytes)
        magic_model = MagicModel(model_list, pdf_docs)
        for i in range(7):
            print(magic_model.get_imgs(i))