CRNN
CRNN(
vocab=DEFAULT_VOCAB,
cnn_channels=(64, 128, 256, 256),
rnn_hidden=256,
rnn_layers=2,
dropout=0.25,
)Convolutional Recurrent Neural Network for variable-length text recognition.
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| vocab | Sequence[str] | DEFAULT_VOCAB |
|
| cnn_channels | Sequence[int] | (64, 128, 256, 256) |
|
| rnn_hidden | int | 256 |
|
| rnn_layers | int | 2 |
|
| dropout | float | 0.25 |
Methods
| Name | Description |
|---|---|
| features | Returns sequence features (B, T, C) ready for the RNN. |
| forward | Returns logits (T, B, num_classes) suitable for CTC loss. |
features
CRNN.features(x)Returns sequence features (B, T, C) ready for the RNN.
forward
CRNN.forward(x)Returns logits (T, B, num_classes) suitable for CTC loss.