rec_postprocess.py 20.5 KB
Newer Older
W
WenmuZhou 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import numpy as np
T
tink2123 已提交
15
import string
W
WenmuZhou 已提交
16 17 18 19 20 21 22 23 24 25 26
import paddle
from paddle.nn import functional as F


class BaseRecLabelDecode(object):
    """ Convert between text-label and text-index """

    def __init__(self,
                 character_dict_path=None,
                 character_type='ch',
                 use_space_char=False):
M
MissPenguin 已提交
27
        support_character_type = [
T
tink2123 已提交
28
            'ch', 'en', 'EN_symbol', 'french', 'german', 'japan', 'korean',
T
tink2123 已提交
29 30
            'it', 'xi', 'pu', 'ru', 'ar', 'ta', 'ug', 'fa', 'ur', 'rs', 'oc',
            'rsc', 'bg', 'uk', 'be', 'te', 'ka', 'chinese_cht', 'hi', 'mr',
T
tink2123 已提交
31
            'ne', 'EN', 'latin', 'arabic', 'cyrillic', 'devanagari'
M
MissPenguin 已提交
32
        ]
W
WenmuZhou 已提交
33
        assert character_type in support_character_type, "Only {} are supported now but get {}".format(
M
MissPenguin 已提交
34
            support_character_type, character_type)
W
WenmuZhou 已提交
35

T
tink2123 已提交
36 37 38
        self.beg_str = "sos"
        self.end_str = "eos"

W
WenmuZhou 已提交
39 40 41
        if character_type == "en":
            self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
            dict_character = list(self.character_str)
T
tink2123 已提交
42
        elif character_type == "EN_symbol":
T
tink2123 已提交
43 44 45 46
            # same with ASTER setting (use 94 char).
            self.character_str = string.printable[:-6]
            dict_character = list(self.character_str)
        elif character_type in support_character_type:
W
WenmuZhou 已提交
47
            self.character_str = []
T
tink2123 已提交
48 49
            assert character_dict_path is not None, "character_dict_path should not be None when character_type is {}".format(
                character_type)
W
WenmuZhou 已提交
50 51 52 53
            with open(character_dict_path, "rb") as fin:
                lines = fin.readlines()
                for line in lines:
                    line = line.decode('utf-8').strip("\n").strip("\r\n")
W
WenmuZhou 已提交
54
                    self.character_str.append(line)
W
WenmuZhou 已提交
55
            if use_space_char:
W
WenmuZhou 已提交
56
                self.character_str.append(" ")
W
WenmuZhou 已提交
57
            dict_character = list(self.character_str)
T
tink2123 已提交
58

W
WenmuZhou 已提交
59 60 61 62 63 64 65 66 67 68 69 70
        else:
            raise NotImplementedError
        self.character_type = character_type
        dict_character = self.add_special_char(dict_character)
        self.dict = {}
        for i, char in enumerate(dict_character):
            self.dict[char] = i
        self.character = dict_character

    def add_special_char(self, dict_character):
        return dict_character

L
littletomatodonkey 已提交
71
    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
W
WenmuZhou 已提交
72 73 74 75 76 77 78 79 80 81 82
        """ convert text-index into text-label. """
        result_list = []
        ignored_tokens = self.get_ignored_tokens()
        batch_size = len(text_index)
        for batch_idx in range(batch_size):
            char_list = []
            conf_list = []
            for idx in range(len(text_index[batch_idx])):
                if text_index[batch_idx][idx] in ignored_tokens:
                    continue
                if is_remove_duplicate:
83
                    # only for predict
W
WenmuZhou 已提交
84 85 86 87 88 89 90 91 92 93
                    if idx > 0 and text_index[batch_idx][idx - 1] == text_index[
                            batch_idx][idx]:
                        continue
                char_list.append(self.character[int(text_index[batch_idx][
                    idx])])
                if text_prob is not None:
                    conf_list.append(text_prob[batch_idx][idx])
                else:
                    conf_list.append(1)
            text = ''.join(char_list)
Z
zhoujun 已提交
94
            result_list.append((text, np.mean(conf_list)))
W
WenmuZhou 已提交
95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
        return result_list

    def get_ignored_tokens(self):
        return [0]  # for ctc blank


class CTCLabelDecode(BaseRecLabelDecode):
    """ Convert between text-label and text-index """

    def __init__(self,
                 character_dict_path=None,
                 character_type='ch',
                 use_space_char=False,
                 **kwargs):
        super(CTCLabelDecode, self).__init__(character_dict_path,
                                             character_type, use_space_char)

    def __call__(self, preds, label=None, *args, **kwargs):
W
WenmuZhou 已提交
113 114
        if isinstance(preds, paddle.Tensor):
            preds = preds.numpy()
W
WenmuZhou 已提交
115 116
        preds_idx = preds.argmax(axis=2)
        preds_prob = preds.max(axis=2)
W
WenmuZhou 已提交
117
        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=True)
W
WenmuZhou 已提交
118 119
        if label is None:
            return text
L
littletomatodonkey 已提交
120
        label = self.decode(label)
W
WenmuZhou 已提交
121 122 123 124 125 126 127
        return text, label

    def add_special_char(self, dict_character):
        dict_character = ['blank'] + dict_character
        return dict_character


littletomatodonkey's avatar
littletomatodonkey 已提交
128 129 130 131 132 133 134 135 136 137
class DistillationCTCLabelDecode(CTCLabelDecode):
    """
    Convert 
    Convert between text-label and text-index
    """

    def __init__(self,
                 character_dict_path=None,
                 character_type='ch',
                 use_space_char=False,
littletomatodonkey's avatar
littletomatodonkey 已提交
138
                 model_name=["student"],
139
                 key=None,
littletomatodonkey's avatar
littletomatodonkey 已提交
140 141 142
                 **kwargs):
        super(DistillationCTCLabelDecode, self).__init__(
            character_dict_path, character_type, use_space_char)
littletomatodonkey's avatar
littletomatodonkey 已提交
143 144
        if not isinstance(model_name, list):
            model_name = [model_name]
littletomatodonkey's avatar
littletomatodonkey 已提交
145
        self.model_name = model_name
littletomatodonkey's avatar
littletomatodonkey 已提交
146

147
        self.key = key
littletomatodonkey's avatar
littletomatodonkey 已提交
148 149

    def __call__(self, preds, label=None, *args, **kwargs):
littletomatodonkey's avatar
littletomatodonkey 已提交
150 151 152 153 154 155 156
        output = dict()
        for name in self.model_name:
            pred = preds[name]
            if self.key is not None:
                pred = pred[self.key]
            output[name] = super().__call__(pred, label=label, *args, **kwargs)
        return output
littletomatodonkey's avatar
littletomatodonkey 已提交
157 158


W
WenmuZhou 已提交
159 160 161 162 163 164 165 166 167 168 169 170
class AttnLabelDecode(BaseRecLabelDecode):
    """ Convert between text-label and text-index """

    def __init__(self,
                 character_dict_path=None,
                 character_type='ch',
                 use_space_char=False,
                 **kwargs):
        super(AttnLabelDecode, self).__init__(character_dict_path,
                                              character_type, use_space_char)

    def add_special_char(self, dict_character):
L
LDOUBLEV 已提交
171 172 173
        self.beg_str = "sos"
        self.end_str = "eos"
        dict_character = dict_character
T
tink2123 已提交
174
        dict_character = [self.beg_str] + dict_character + [self.end_str]
W
WenmuZhou 已提交
175 176
        return dict_character

L
LDOUBLEV 已提交
177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205
    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
        """ convert text-index into text-label. """
        result_list = []
        ignored_tokens = self.get_ignored_tokens()
        [beg_idx, end_idx] = self.get_ignored_tokens()
        batch_size = len(text_index)
        for batch_idx in range(batch_size):
            char_list = []
            conf_list = []
            for idx in range(len(text_index[batch_idx])):
                if text_index[batch_idx][idx] in ignored_tokens:
                    continue
                if int(text_index[batch_idx][idx]) == int(end_idx):
                    break
                if is_remove_duplicate:
                    # only for predict
                    if idx > 0 and text_index[batch_idx][idx - 1] == text_index[
                            batch_idx][idx]:
                        continue
                char_list.append(self.character[int(text_index[batch_idx][
                    idx])])
                if text_prob is not None:
                    conf_list.append(text_prob[batch_idx][idx])
                else:
                    conf_list.append(1)
            text = ''.join(char_list)
            result_list.append((text, np.mean(conf_list)))
        return result_list

L
LDOUBLEV 已提交
206 207
    def __call__(self, preds, label=None, *args, **kwargs):
        """
W
WenmuZhou 已提交
208
        text = self.decode(text)
L
LDOUBLEV 已提交
209 210 211 212 213 214 215 216 217 218 219
        if label is None:
            return text
        else:
            label = self.decode(label, is_remove_duplicate=False)
            return text, label
        """
        if isinstance(preds, paddle.Tensor):
            preds = preds.numpy()

        preds_idx = preds.argmax(axis=2)
        preds_prob = preds.max(axis=2)
L
LDOUBLEV 已提交
220
        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
L
LDOUBLEV 已提交
221 222
        if label is None:
            return text
L
LDOUBLEV 已提交
223
        label = self.decode(label, is_remove_duplicate=False)
L
LDOUBLEV 已提交
224 225
        return text, label

W
WenmuZhou 已提交
226 227 228 229 230 231 232 233 234 235 236 237 238
    def get_ignored_tokens(self):
        beg_idx = self.get_beg_end_flag_idx("beg")
        end_idx = self.get_beg_end_flag_idx("end")
        return [beg_idx, end_idx]

    def get_beg_end_flag_idx(self, beg_or_end):
        if beg_or_end == "beg":
            idx = np.array(self.dict[self.beg_str])
        elif beg_or_end == "end":
            idx = np.array(self.dict[self.end_str])
        else:
            assert False, "unsupport type %s in get_beg_end_flag_idx" \
                          % beg_or_end
M
MissPenguin 已提交
239
        return idx
T
tink2123 已提交
240 241


T
tink2123 已提交
242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323
class SEEDLabelDecode(BaseRecLabelDecode):
    """ Convert between text-label and text-index """

    def __init__(self,
                 character_dict_path=None,
                 character_type='ch',
                 use_space_char=False,
                 **kwargs):
        super(SEEDLabelDecode, self).__init__(character_dict_path,
                                              character_type, use_space_char)

    def add_special_char(self, dict_character):
        self.beg_str = "sos"
        self.end_str = "eos"
        dict_character = dict_character
        dict_character = dict_character + [self.end_str]
        return dict_character

    def get_ignored_tokens(self):
        end_idx = self.get_beg_end_flag_idx("eos")
        return [end_idx]

    def get_beg_end_flag_idx(self, beg_or_end):
        if beg_or_end == "sos":
            idx = np.array(self.dict[self.beg_str])
        elif beg_or_end == "eos":
            idx = np.array(self.dict[self.end_str])
        else:
            assert False, "unsupport type %s in get_beg_end_flag_idx" % beg_or_end
        return idx

    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
        """ convert text-index into text-label. """
        result_list = []
        [end_idx] = self.get_ignored_tokens()
        batch_size = len(text_index)
        for batch_idx in range(batch_size):
            char_list = []
            conf_list = []
            for idx in range(len(text_index[batch_idx])):
                if int(text_index[batch_idx][idx]) == int(end_idx):
                    break
                if is_remove_duplicate:
                    # only for predict
                    if idx > 0 and text_index[batch_idx][idx - 1] == text_index[
                            batch_idx][idx]:
                        continue
                char_list.append(self.character[int(text_index[batch_idx][
                    idx])])
                if text_prob is not None:
                    conf_list.append(text_prob[batch_idx][idx])
                else:
                    conf_list.append(1)
            text = ''.join(char_list)
            result_list.append((text, np.mean(conf_list)))
        return result_list

    def __call__(self, preds, label=None, *args, **kwargs):
        """
        text = self.decode(text)
        if label is None:
            return text
        else:
            label = self.decode(label, is_remove_duplicate=False)
            return text, label
        """
        preds_idx = preds["rec_pred"]
        if isinstance(preds_idx, paddle.Tensor):
            preds_idx = preds_idx.numpy()
        if "rec_pred_scores" in preds:
            preds_idx = preds["rec_pred"]
            preds_prob = preds["rec_pred_scores"]
        else:
            preds_idx = preds["rec_pred"].argmax(axis=2)
            preds_prob = preds["rec_pred"].max(axis=2)
        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
        if label is None:
            return text
        label = self.decode(label, is_remove_duplicate=False)
        return text, label


T
tink2123 已提交
324 325 326 327 328 329 330 331 332 333
class SRNLabelDecode(BaseRecLabelDecode):
    """ Convert between text-label and text-index """

    def __init__(self,
                 character_dict_path=None,
                 character_type='en',
                 use_space_char=False,
                 **kwargs):
        super(SRNLabelDecode, self).__init__(character_dict_path,
                                             character_type, use_space_char)
334
        self.max_text_length = kwargs.get('max_text_length', 25)
T
tink2123 已提交
335 336 337 338 339 340 341 342 343 344 345

    def __call__(self, preds, label=None, *args, **kwargs):
        pred = preds['predict']
        char_num = len(self.character_str) + 2
        if isinstance(pred, paddle.Tensor):
            pred = pred.numpy()
        pred = np.reshape(pred, [-1, char_num])

        preds_idx = np.argmax(pred, axis=1)
        preds_prob = np.max(pred, axis=1)

346
        preds_idx = np.reshape(preds_idx, [-1, self.max_text_length])
T
tink2123 已提交
347

348
        preds_prob = np.reshape(preds_prob, [-1, self.max_text_length])
T
tink2123 已提交
349

T
tink2123 已提交
350
        text = self.decode(preds_idx, preds_prob)
T
tink2123 已提交
351 352

        if label is None:
L
LDOUBLEV 已提交
353
            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
T
tink2123 已提交
354
            return text
T
tink2123 已提交
355
        label = self.decode(label)
T
tink2123 已提交
356 357
        return text, label

L
LDOUBLEV 已提交
358
    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
T
tink2123 已提交
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403
        """ convert text-index into text-label. """
        result_list = []
        ignored_tokens = self.get_ignored_tokens()
        batch_size = len(text_index)

        for batch_idx in range(batch_size):
            char_list = []
            conf_list = []
            for idx in range(len(text_index[batch_idx])):
                if text_index[batch_idx][idx] in ignored_tokens:
                    continue
                if is_remove_duplicate:
                    # only for predict
                    if idx > 0 and text_index[batch_idx][idx - 1] == text_index[
                            batch_idx][idx]:
                        continue
                char_list.append(self.character[int(text_index[batch_idx][
                    idx])])
                if text_prob is not None:
                    conf_list.append(text_prob[batch_idx][idx])
                else:
                    conf_list.append(1)

            text = ''.join(char_list)
            result_list.append((text, np.mean(conf_list)))
        return result_list

    def add_special_char(self, dict_character):
        dict_character = dict_character + [self.beg_str, self.end_str]
        return dict_character

    def get_ignored_tokens(self):
        beg_idx = self.get_beg_end_flag_idx("beg")
        end_idx = self.get_beg_end_flag_idx("end")
        return [beg_idx, end_idx]

    def get_beg_end_flag_idx(self, beg_or_end):
        if beg_or_end == "beg":
            idx = np.array(self.dict[self.beg_str])
        elif beg_or_end == "end":
            idx = np.array(self.dict[self.end_str])
        else:
            assert False, "unsupport type %s in get_beg_end_flag_idx" \
                          % beg_or_end
        return idx
W
WenmuZhou 已提交
404 405 406 407 408


class TableLabelDecode(object):
    """  """

T
tink2123 已提交
409 410 411
    def __init__(self, character_dict_path, **kwargs):
        list_character, list_elem = self.load_char_elem_dict(
            character_dict_path)
W
WenmuZhou 已提交
412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449
        list_character = self.add_special_char(list_character)
        list_elem = self.add_special_char(list_elem)
        self.dict_character = {}
        self.dict_idx_character = {}
        for i, char in enumerate(list_character):
            self.dict_idx_character[i] = char
            self.dict_character[char] = i
        self.dict_elem = {}
        self.dict_idx_elem = {}
        for i, elem in enumerate(list_elem):
            self.dict_idx_elem[i] = elem
            self.dict_elem[elem] = i

    def load_char_elem_dict(self, character_dict_path):
        list_character = []
        list_elem = []
        with open(character_dict_path, "rb") as fin:
            lines = fin.readlines()
            substr = lines[0].decode('utf-8').strip("\n").split("\t")
            character_num = int(substr[0])
            elem_num = int(substr[1])
            for cno in range(1, 1 + character_num):
                character = lines[cno].decode('utf-8').strip("\n")
                list_character.append(character)
            for eno in range(1 + character_num, 1 + character_num + elem_num):
                elem = lines[eno].decode('utf-8').strip("\n")
                list_elem.append(elem)
        return list_character, list_elem

    def add_special_char(self, list_character):
        self.beg_str = "sos"
        self.end_str = "eos"
        list_character = [self.beg_str] + list_character + [self.end_str]
        return list_character

    def __call__(self, preds):
        structure_probs = preds['structure_probs']
        loc_preds = preds['loc_preds']
T
tink2123 已提交
450
        if isinstance(structure_probs, paddle.Tensor):
W
WenmuZhou 已提交
451
            structure_probs = structure_probs.numpy()
T
tink2123 已提交
452
        if isinstance(loc_preds, paddle.Tensor):
W
WenmuZhou 已提交
453 454 455
            loc_preds = loc_preds.numpy()
        structure_idx = structure_probs.argmax(axis=2)
        structure_probs = structure_probs.max(axis=2)
T
tink2123 已提交
456 457
        structure_str, structure_pos, result_score_list, result_elem_idx_list = self.decode(
            structure_idx, structure_probs, 'elem')
W
WenmuZhou 已提交
458 459 460 461 462 463 464 465 466 467 468 469 470 471
        res_html_code_list = []
        res_loc_list = []
        batch_num = len(structure_str)
        for bno in range(batch_num):
            res_loc = []
            for sno in range(len(structure_str[bno])):
                text = structure_str[bno][sno]
                if text in ['<td>', '<td']:
                    pos = structure_pos[bno][sno]
                    res_loc.append(loc_preds[bno, pos])
            res_html_code = ''.join(structure_str[bno])
            res_loc = np.array(res_loc)
            res_html_code_list.append(res_html_code)
            res_loc_list.append(res_loc)
T
tink2123 已提交
472 473 474 475 476 477 478
        return {
            'res_html_code': res_html_code_list,
            'res_loc': res_loc_list,
            'res_score_list': result_score_list,
            'res_elem_idx_list': result_elem_idx_list,
            'structure_str_list': structure_str
        }
W
WenmuZhou 已提交
479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542

    def decode(self, text_index, structure_probs, char_or_elem):
        """convert text-label into text-index.
        """
        if char_or_elem == "char":
            current_dict = self.dict_idx_character
        else:
            current_dict = self.dict_idx_elem
            ignored_tokens = self.get_ignored_tokens('elem')
            beg_idx, end_idx = ignored_tokens

        result_list = []
        result_pos_list = []
        result_score_list = []
        result_elem_idx_list = []
        batch_size = len(text_index)
        for batch_idx in range(batch_size):
            char_list = []
            elem_pos_list = []
            elem_idx_list = []
            score_list = []
            for idx in range(len(text_index[batch_idx])):
                tmp_elem_idx = int(text_index[batch_idx][idx])
                if idx > 0 and tmp_elem_idx == end_idx:
                    break
                if tmp_elem_idx in ignored_tokens:
                    continue

                char_list.append(current_dict[tmp_elem_idx])
                elem_pos_list.append(idx)
                score_list.append(structure_probs[batch_idx, idx])
                elem_idx_list.append(tmp_elem_idx)
            result_list.append(char_list)
            result_pos_list.append(elem_pos_list)
            result_score_list.append(score_list)
            result_elem_idx_list.append(elem_idx_list)
        return result_list, result_pos_list, result_score_list, result_elem_idx_list

    def get_ignored_tokens(self, char_or_elem):
        beg_idx = self.get_beg_end_flag_idx("beg", char_or_elem)
        end_idx = self.get_beg_end_flag_idx("end", char_or_elem)
        return [beg_idx, end_idx]

    def get_beg_end_flag_idx(self, beg_or_end, char_or_elem):
        if char_or_elem == "char":
            if beg_or_end == "beg":
                idx = self.dict_character[self.beg_str]
            elif beg_or_end == "end":
                idx = self.dict_character[self.end_str]
            else:
                assert False, "Unsupport type %s in get_beg_end_flag_idx of char" \
                              % beg_or_end
        elif char_or_elem == "elem":
            if beg_or_end == "beg":
                idx = self.dict_elem[self.beg_str]
            elif beg_or_end == "end":
                idx = self.dict_elem[self.end_str]
            else:
                assert False, "Unsupport type %s in get_beg_end_flag_idx of elem" \
                              % beg_or_end
        else:
            assert False, "Unsupport type %s in char_or_elem" \
                          % char_or_elem
        return idx