Spaces:
Runtime error
Runtime error
File size: 6,121 Bytes
2366e36 |
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 |
# Copyright (c) OpenMMLab. All rights reserved.
import warnings
import torch
from mmocr.models.builder import (RECOGNIZERS, build_backbone, build_convertor,
build_decoder, build_encoder, build_loss,
build_preprocessor)
from .base import BaseRecognizer
@RECOGNIZERS.register_module()
class EncodeDecodeRecognizer(BaseRecognizer):
"""Base class for encode-decode recognizer."""
def __init__(self,
preprocessor=None,
backbone=None,
encoder=None,
decoder=None,
loss=None,
label_convertor=None,
train_cfg=None,
test_cfg=None,
max_seq_len=40,
pretrained=None,
init_cfg=None):
super().__init__(init_cfg=init_cfg)
# Label convertor (str2tensor, tensor2str)
assert label_convertor is not None
label_convertor.update(max_seq_len=max_seq_len)
self.label_convertor = build_convertor(label_convertor)
# Preprocessor module, e.g., TPS
self.preprocessor = None
if preprocessor is not None:
self.preprocessor = build_preprocessor(preprocessor)
# Backbone
assert backbone is not None
self.backbone = build_backbone(backbone)
# Encoder module
self.encoder = None
if encoder is not None:
self.encoder = build_encoder(encoder)
# Decoder module
assert decoder is not None
decoder.update(num_classes=self.label_convertor.num_classes())
decoder.update(start_idx=self.label_convertor.start_idx)
decoder.update(padding_idx=self.label_convertor.padding_idx)
decoder.update(max_seq_len=max_seq_len)
self.decoder = build_decoder(decoder)
# Loss
assert loss is not None
loss.update(ignore_index=self.label_convertor.padding_idx)
self.loss = build_loss(loss)
self.train_cfg = train_cfg
self.test_cfg = test_cfg
self.max_seq_len = max_seq_len
if pretrained is not None:
warnings.warn('DeprecationWarning: pretrained is a deprecated \
key, please consider using init_cfg')
self.init_cfg = dict(type='Pretrained', checkpoint=pretrained)
def extract_feat(self, img):
"""Directly extract features from the backbone."""
if self.preprocessor is not None:
img = self.preprocessor(img)
x = self.backbone(img)
return x
def forward_train(self, img, img_metas):
"""
Args:
img (tensor): Input images of shape (N, C, H, W).
Typically these should be mean centered and std scaled.
img_metas (list[dict]): A list of image info dict where each dict
contains: 'img_shape', 'filename', and may also contain
'ori_shape', and 'img_norm_cfg'.
For details on the values of these keys see
:class:`mmdet.datasets.pipelines.Collect`.
Returns:
dict[str, tensor]: A dictionary of loss components.
"""
for img_meta in img_metas:
valid_ratio = 1.0 * img_meta['resize_shape'][1] / img.size(-1)
img_meta['valid_ratio'] = valid_ratio
feat = self.extract_feat(img)
gt_labels = [img_meta['text'] for img_meta in img_metas]
targets_dict = self.label_convertor.str2tensor(gt_labels)
out_enc = None
if self.encoder is not None:
out_enc = self.encoder(feat, img_metas)
out_dec = self.decoder(
feat, out_enc, targets_dict, img_metas, train_mode=True)
loss_inputs = (
out_dec,
targets_dict,
img_metas,
)
losses = self.loss(*loss_inputs)
return losses
def simple_test(self, img, img_metas, **kwargs):
"""Test function with test time augmentation.
Args:
imgs (torch.Tensor): Image input tensor.
img_metas (list[dict]): List of image information.
Returns:
list[str]: Text label result of each image.
"""
for img_meta in img_metas:
valid_ratio = 1.0 * img_meta['resize_shape'][1] / img.size(-1)
img_meta['valid_ratio'] = valid_ratio
feat = self.extract_feat(img)
out_enc = None
if self.encoder is not None:
out_enc = self.encoder(feat, img_metas)
out_dec = self.decoder(
feat, out_enc, None, img_metas, train_mode=False)
# early return to avoid post processing
if torch.onnx.is_in_onnx_export():
return out_dec
label_indexes, label_scores = self.label_convertor.tensor2idx(
out_dec, img_metas)
label_strings = self.label_convertor.idx2str(label_indexes)
# flatten batch results
results = []
for string, score in zip(label_strings, label_scores):
results.append(dict(text=string, score=score))
return results
def merge_aug_results(self, aug_results):
out_text, out_score = '', -1
for result in aug_results:
text = result[0]['text']
score = sum(result[0]['score']) / max(1, len(text))
if score > out_score:
out_text = text
out_score = score
out_results = [dict(text=out_text, score=out_score)]
return out_results
def aug_test(self, imgs, img_metas, **kwargs):
"""Test function as well as time augmentation.
Args:
imgs (list[tensor]): Tensor should have shape NxCxHxW,
which contains all images in the batch.
img_metas (list[list[dict]]): The metadata of images.
"""
aug_results = []
for img, img_meta in zip(imgs, img_metas):
result = self.simple_test(img, img_meta, **kwargs)
aug_results.append(result)
return self.merge_aug_results(aug_results)
|