File size: 6,829 Bytes
a89d9fd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
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
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# 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
import paddle

from .rec_postprocess import AttnLabelDecode


class TableLabelDecode(AttnLabelDecode):
    """  """

    def __init__(self,
                 character_dict_path,
                 merge_no_span_structure=False,
                 **kwargs):
        dict_character = []
        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")
                dict_character.append(line)

        if merge_no_span_structure:
            if "<td></td>" not in dict_character:
                dict_character.append("<td></td>")
            if "<td>" in dict_character:
                dict_character.remove("<td>")

        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
        self.td_token = ['<td>', '<td', '<td></td>']

    def __call__(self, preds, batch=None):
        structure_probs = preds['structure_probs']
        bbox_preds = preds['loc_preds']
        if isinstance(structure_probs, paddle.Tensor):
            structure_probs = structure_probs.numpy()
        if isinstance(bbox_preds, paddle.Tensor):
            bbox_preds = bbox_preds.numpy()
        shape_list = batch[-1]
        result = self.decode(structure_probs, bbox_preds, shape_list)
        if len(batch) == 1:  # only contains shape
            return result

        label_decode_result = self.decode_label(batch)
        return result, label_decode_result

    def decode(self, structure_probs, bbox_preds, shape_list):
        """convert text-label into text-index.
        """
        ignored_tokens = self.get_ignored_tokens()
        end_idx = self.dict[self.end_str]

        structure_idx = structure_probs.argmax(axis=2)
        structure_probs = structure_probs.max(axis=2)

        structure_batch_list = []
        bbox_batch_list = []
        batch_size = len(structure_idx)
        for batch_idx in range(batch_size):
            structure_list = []
            bbox_list = []
            score_list = []
            for idx in range(len(structure_idx[batch_idx])):
                char_idx = int(structure_idx[batch_idx][idx])
                if idx > 0 and char_idx == end_idx:
                    break
                if char_idx in ignored_tokens:
                    continue
                text = self.character[char_idx]
                if text in self.td_token:
                    bbox = bbox_preds[batch_idx, idx]
                    bbox = self._bbox_decode(bbox, shape_list[batch_idx])
                    bbox_list.append(bbox)
                structure_list.append(text)
                score_list.append(structure_probs[batch_idx, idx])
            structure_batch_list.append([structure_list, np.mean(score_list)])
            bbox_batch_list.append(np.array(bbox_list))
        result = {
            'bbox_batch_list': bbox_batch_list,
            'structure_batch_list': structure_batch_list,
        }
        return result

    def decode_label(self, batch):
        """convert text-label into text-index.
        """
        structure_idx = batch[1]
        gt_bbox_list = batch[2]
        shape_list = batch[-1]
        ignored_tokens = self.get_ignored_tokens()
        end_idx = self.dict[self.end_str]

        structure_batch_list = []
        bbox_batch_list = []
        batch_size = len(structure_idx)
        for batch_idx in range(batch_size):
            structure_list = []
            bbox_list = []
            for idx in range(len(structure_idx[batch_idx])):
                char_idx = int(structure_idx[batch_idx][idx])
                if idx > 0 and char_idx == end_idx:
                    break
                if char_idx in ignored_tokens:
                    continue
                structure_list.append(self.character[char_idx])

                bbox = gt_bbox_list[batch_idx][idx]
                if bbox.sum() != 0:
                    bbox = self._bbox_decode(bbox, shape_list[batch_idx])
                    bbox_list.append(bbox)
            structure_batch_list.append(structure_list)
            bbox_batch_list.append(bbox_list)
        result = {
            'bbox_batch_list': bbox_batch_list,
            'structure_batch_list': structure_batch_list,
        }
        return result

    def _bbox_decode(self, bbox, shape):
        h, w, ratio_h, ratio_w, pad_h, pad_w = shape
        bbox[0::2] *= w
        bbox[1::2] *= h
        return bbox


class TableMasterLabelDecode(TableLabelDecode):
    """  """

    def __init__(self,
                 character_dict_path,
                 box_shape='ori',
                 merge_no_span_structure=True,
                 **kwargs):
        super(TableMasterLabelDecode, self).__init__(character_dict_path,
                                                     merge_no_span_structure)
        self.box_shape = box_shape
        assert box_shape in [
            'ori', 'pad'
        ], 'The shape used for box normalization must be ori or pad'

    def add_special_char(self, dict_character):
        self.beg_str = '<SOS>'
        self.end_str = '<EOS>'
        self.unknown_str = '<UKN>'
        self.pad_str = '<PAD>'
        dict_character = dict_character
        dict_character = dict_character + [
            self.unknown_str, self.beg_str, self.end_str, self.pad_str
        ]
        return dict_character

    def get_ignored_tokens(self):
        pad_idx = self.dict[self.pad_str]
        start_idx = self.dict[self.beg_str]
        end_idx = self.dict[self.end_str]
        unknown_idx = self.dict[self.unknown_str]
        return [start_idx, end_idx, pad_idx, unknown_idx]

    def _bbox_decode(self, bbox, shape):
        h, w, ratio_h, ratio_w, pad_h, pad_w = shape
        if self.box_shape == 'pad':
            h, w = pad_h, pad_w
        bbox[0::2] *= w
        bbox[1::2] *= h
        bbox[0::2] /= ratio_w
        bbox[1::2] /= ratio_h
        x, y, w, h = bbox
        x1, y1, x2, y2 = x - w // 2, y - h // 2, x + w // 2, y + h // 2
        bbox = np.array([x1, y1, x2, y2])
        return bbox