File size: 6,037 Bytes
cb0ad2d |
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 |
import os
import copy
import json
import pickle
import random
from torch._C import layout
import tqdm
import torch
import numpy as np
from PIL import Image
from .list_record_cache import ListRecordLoader
from libs.utils.format_translate import table_to_html
class LRCRecordLoader:
def __init__(self, lrc_path, data_dir=''):
self.loader = ListRecordLoader(lrc_path)
self.data_root_dir = data_dir
def __len__(self):
return len(self.loader)
def get_info(self, idx):
table = self.loader.get_record(idx)
image = Image.open(table['image_path']).convert('RGB')
w = image.width
h = image.height
n_rows, n_cols = table['layout'].shape
n_cells = n_rows * n_cols
return w, h, n_cells
def get_data(self, idx):
table = self.loader.get_record(idx)
img_path = os.path.join(self.data_root_dir, table['image_path'])
image = Image.open(img_path).convert('RGB')
return image, table
class Dataset:
def __init__(self, loaders, transforms):
self.loaders = loaders
self.transforms = transforms
def _match_loader(self, idx):
offset = 0
for loader in self.loaders:
if len(loader) + offset > idx:
return loader, idx - offset
else:
offset += len(loader)
raise IndexError()
def get_info(self, idx):
loader, rela_idx = self._match_loader(idx)
return loader.get_info(rela_idx)
def __len__(self):
return sum([len(loader) for loader in self.loaders])
def __getitem__(self,idx):
try:
loader, rela_idx = self._match_loader(idx)
image, table = loader.get_data(rela_idx)
image, _, cls_label, \
rows_fg_span, rows_bg_span, \
cols_fg_span, cols_bg_span, \
cells_span, divide = self.transforms(image, table) if 'layout' in table.keys() else self.transforms(image)
return dict(
id=idx,
image_size=(image.shape[2], image.shape[1]),
image=image,
cls_label=cls_label,
rows_fg_span=rows_fg_span,
rows_bg_span=rows_bg_span,
cols_fg_span=cols_fg_span,
cols_bg_span=cols_bg_span,
cells_span=cells_span,
layout=table['layout'] if 'layout' in table.keys() else None,
divide=divide,
table=table
)
except Exception as e:
print('Error occured while load data: %d' % idx)
raise e
def collate_func(batch_data):
batch_size = len(batch_data)
image_dim = batch_data[0]['image'].shape[0]
max_h = max([data['image'].shape[1] for data in batch_data])
max_w = max([data['image'].shape[2] for data in batch_data])
batch_id = list()
batch_image_size = list()
batch_image = torch.zeros([batch_size, image_dim, max_h, max_w], dtype=torch.float)
batch_image_mask = torch.zeros([batch_size, 1, max_h, max_w], dtype=torch.float)
batch_rows_fg_span = list()
batch_rows_bg_span = list()
batch_cols_fg_span = list()
batch_cols_bg_span = list()
batch_cells_span = list()
batch_divide = list()
tables = list()
if all([(data['cls_label'] is None) and (data['layout'] is None) for data in batch_data]):
batch_cls_label = list()
batch_label_mask = list()
batch_layout = list()
else:
assert not any([(data['cls_label'] is None) or (data['layout'] is None) for data in batch_data])
max_label_length = max([data['cls_label'].shape[0] for data in batch_data])
batch_cls_label = torch.zeros([batch_size, max_label_length], dtype=torch.long)
batch_label_mask = torch.zeros([batch_size, max_label_length], dtype=torch.float)
max_nr = max([data['layout'].shape[0] for data in batch_data])
max_nc = max([data['layout'].shape[1] for data in batch_data])
batch_layout = torch.full([batch_size, max_nr, max_nc], -1, dtype=torch.float)
for batch_idx, data in enumerate(batch_data):
batch_id.append(data['id'])
batch_image_size.append(data['image_size'])
_, cur_h, cur_w = data['image'].shape
batch_image[batch_idx, :, :cur_h, :cur_w] = data["image"]
batch_image_mask[batch_idx, :, :cur_h, :cur_w] = 1
if (data['cls_label'] is None) and (data['layout'] is None):
batch_cls_label.append(data["cls_label"])
batch_label_mask.append(None)
batch_layout.append(data["layout"])
else:
label_length = data['cls_label'].shape[0]
batch_cls_label[batch_idx, :label_length] = data['cls_label']
batch_label_mask[batch_idx, :label_length] = 1.0
layout_nr, layout_nc = data["layout" ].shape
batch_layout[batch_idx, :layout_nr, :layout_nc] = torch.from_numpy(data['layout']).float()
batch_rows_fg_span.append(data["rows_fg_span"])
batch_rows_bg_span.append(data['rows_bg_span'])
batch_cols_fg_span.append(data["cols_fg_span"])
batch_cols_bg_span.append(data["cols_bg_span"])
batch_cells_span.append(data["cells_span"])
batch_divide.append(data["divide"])
tables.append(data['table'])
batch_divide = torch.tensor(batch_divide, dtype=torch.long) if batch_divide[0] is not None else batch_divide
return dict(
ids=batch_id,
images_size=batch_image_size,
images=batch_image,
images_mask=batch_image_mask,
cls_labels=batch_cls_label,
labels_mask=batch_label_mask,
rows_fg_spans=batch_rows_fg_span,
rows_bg_spans=batch_rows_bg_span,
cols_fg_spans=batch_cols_fg_span,
cols_bg_spans=batch_cols_bg_span,
cells_spans=batch_cells_span,
divide_labels=batch_divide,
layouts=batch_layout,
tables=tables
) |