Add vq transformer network

2 files changed, 58 insertions, 0 deletions

diff --git a/text_recognizer/networks/__init__.py b/text_recognizer/networks/__init__.py
index dad9d23..e2d6fd5 100644
--- a/text_recognizer/networks/__init__.py
+++ b/text_recognizer/networks/__init__.py
@@ -1,3 +1,4 @@
 """Network modules"""
 from text_recognizer.networks.conv_transformer import ConvTransformer
 from text_recognizer.networks.efficientnet.efficientnet import EfficientNet
+from text_recognizer.networks.vq_transformer import VqTransformer
diff --git a/text_recognizer/networks/vq_transformer.py b/text_recognizer/networks/vq_transformer.py
new file mode 100644
index 0000000..c12e18b
--- /dev/null
+++ b/text_recognizer/networks/vq_transformer.py
@@ -0,0 +1,57 @@
+from typing import Optional, Tuple, Type
+
+from torch import nn, Tensor
+
+from text_recognizer.networks.transformer.decoder import Decoder
+from text_recognizer.networks.transformer.embeddings.axial import (
+    AxialPositionalEmbedding,
+)
+
+from text_recognizer.networks.conv_transformer import ConvTransformer
+from text_recognizer.networks.quantizer.quantizer import VectorQuantizer
+
+
+class VqTransformer(ConvTransformer):
+    def __init__(
+        self,
+        input_dims: Tuple[int, int, int],
+        hidden_dim: int,
+        num_classes: int,
+        pad_index: Tensor,
+        encoder: Type[nn.Module],
+        decoder: Decoder,
+        pixel_embedding: AxialPositionalEmbedding,
+        token_pos_embedding: Optional[Type[nn.Module]] = None,
+        quantizer: Optional[VectorQuantizer] = None,
+    ) -> None:
+        super().__init__(
+            input_dims,
+            hidden_dim,
+            num_classes,
+            pad_index,
+            encoder,
+            decoder,
+            pixel_embedding,
+            token_pos_embedding,
+        )
+        self.quantizer = quantizer
+
+    def quantize(self, z: Tensor) -> Tuple[Tensor, Tensor]:
+        q, _, loss = self.quantizer(z)
+        return q, loss
+
+    def encode(self, x: Tensor) -> Tuple[Tensor, Tensor]:
+        z = self.encoder(x)
+        z = self.conv(z)
+        q, loss = self.quantize(z)
+        z = self.pixel_embedding(q)
+        z = z.flatten(start_dim=2)
+
+        # Permute tensor from [B, E, Ho * Wo] to [B, Sx, E]
+        z = z.permute(0, 2, 1)
+        return z, loss
+
+    def forward(self, x: Tensor, context: Tensor) -> Tuple[Tensor, Tensor]:
+        z, loss = self.encode(x)
+        logits = self.decode(z, context)
+        return logits, loss