diff options
| -rw-r--r-- | text_recognizer/networks/efficientnet/efficientnet.py | 2 | ||||
| -rw-r--r-- | training/conf/network/conv_transformer.yaml | 5 |
2 files changed, 2 insertions, 5 deletions
diff --git a/text_recognizer/networks/efficientnet/efficientnet.py b/text_recognizer/networks/efficientnet/efficientnet.py index 2a712d8..c94c748 100644 --- a/text_recognizer/networks/efficientnet/efficientnet.py +++ b/text_recognizer/networks/efficientnet/efficientnet.py @@ -36,7 +36,6 @@ class EfficientNet(nn.Module): bn_eps: float = 1.0e-3, depth: int = 7, out_channels: int = 1280, - stride: Tuple[int, int] = (2, 2), ) -> None: super().__init__() self.params = self._get_arch_params(arch) @@ -44,7 +43,6 @@ class EfficientNet(nn.Module): self.bn_momentum = bn_momentum self.bn_eps = bn_eps self.depth = depth - self.stride = stride self.out_channels: int = out_channels self._conv_stem: nn.Sequential self._blocks: nn.ModuleList diff --git a/training/conf/network/conv_transformer.yaml b/training/conf/network/conv_transformer.yaml index 90c2cb8..016416c 100644 --- a/training/conf/network/conv_transformer.yaml +++ b/training/conf/network/conv_transformer.yaml @@ -1,6 +1,6 @@ _target_: text_recognizer.networks.ConvTransformer input_dims: [1, 1, 576, 640] -hidden_dim: &hidden_dim 144 +hidden_dim: &hidden_dim 128 num_classes: 58 pad_index: 3 encoder: @@ -11,7 +11,6 @@ encoder: bn_eps: 1.0e-3 depth: 5 out_channels: *hidden_dim - stride: [2, 1] decoder: _target_: text_recognizer.networks.transformer.Decoder depth: 6 @@ -47,4 +46,4 @@ decoder: pixel_embedding: _target_: text_recognizer.networks.transformer.AxialPositionalEmbedding dim: *hidden_dim - shape: [18, 78] + shape: [18, 80] |