_target_: text_recognizer.networks.ConvPerceiver
input_dims: [1, 1, 576, 640]
hidden_dim: &hidden_dim 128
num_classes: &num_classes 58
max_length: &max_length 89
num_queries: *max_length
queries_dim: &queries_dim 64
pad_index: 3
encoder:
  _target_: text_recognizer.networks.EfficientNet
  arch: b0
  stochastic_dropout_rate: 0.2
  bn_momentum: 0.99
  bn_eps: 1.0e-3
  depth: 5
  out_channels: *hidden_dim
decoder:
  _target_: text_recognizer.networks.perceiver.PerceiverIO
  dim: 192
  cross_heads: 1
  cross_head_dim: 64
  num_latents: 256
  latent_dim: 512
  latent_heads: 8
  depth: 6
  queries_dim: 128
  logits_dim: *num_classes
pixel_embedding:
  _target_: text_recognizer.networks.transformer.embeddings.axial.AxialPositionalEmbeddingImage
  dim: 64
  axial_shape: [3, 64]
  axial_dims: [32, 32]
query_pos_emb:
  _target_: text_recognizer.networks.transformer.embeddings.absolute.AbsolutePositionalEmbedding
  dim: 64
  max_seq_len: *max_length
  l2norm_embed: true