defaults:
  - encoder: efficientnet
  - decoder: transformer_decoder

_target_: text_recognizer.networks.conv_transformer.ConvTransformer
input_dims: [1, 576, 640]
hidden_dim: 256
dropout_rate: 0.2
max_output_len: 682
num_classes: 1004
start_token: <s>
end_token: <e>
pad_token: <p>