Spaces:
Runtime error
Runtime error
File size: 2,360 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 |
# Copyright (c) OpenMMLab. All rights reserved.
import torch.nn as nn
from mmcv.runner import Sequential
from mmocr.models.builder import DECODERS
from mmocr.models.textrecog.layers import BidirectionalLSTM
from .base_decoder import BaseDecoder
@DECODERS.register_module()
class CRNNDecoder(BaseDecoder):
"""Decoder for CRNN.
Args:
in_channels (int): Number of input channels.
num_classes (int): Number of output classes.
rnn_flag (bool): Use RNN or CNN as the decoder.
init_cfg (dict or list[dict], optional): Initialization configs.
"""
def __init__(self,
in_channels=None,
num_classes=None,
rnn_flag=False,
init_cfg=dict(type='Xavier', layer='Conv2d'),
**kwargs):
super().__init__(init_cfg=init_cfg)
self.num_classes = num_classes
self.rnn_flag = rnn_flag
if rnn_flag:
self.decoder = Sequential(
BidirectionalLSTM(in_channels, 256, 256),
BidirectionalLSTM(256, 256, num_classes))
else:
self.decoder = nn.Conv2d(
in_channels, num_classes, kernel_size=1, stride=1)
def forward_train(self, feat, out_enc, targets_dict, img_metas):
"""
Args:
feat (Tensor): A Tensor of shape :math:`(N, H, 1, W)`.
Returns:
Tensor: The raw logit tensor. Shape :math:`(N, W, C)` where
:math:`C` is ``num_classes``.
"""
assert feat.size(2) == 1, 'feature height must be 1'
if self.rnn_flag:
x = feat.squeeze(2) # [N, C, W]
x = x.permute(2, 0, 1) # [W, N, C]
x = self.decoder(x) # [W, N, C]
outputs = x.permute(1, 0, 2).contiguous()
else:
x = self.decoder(feat)
x = x.permute(0, 3, 1, 2).contiguous()
n, w, c, h = x.size()
outputs = x.view(n, w, c * h)
return outputs
def forward_test(self, feat, out_enc, img_metas):
"""
Args:
feat (Tensor): A Tensor of shape :math:`(N, H, 1, W)`.
Returns:
Tensor: The raw logit tensor. Shape :math:`(N, W, C)` where
:math:`C` is ``num_classes``.
"""
return self.forward_train(feat, out_enc, None, img_metas)
|