summaryrefslogtreecommitdiff
path: root/text_recognizer/networks/text_decoder.py
diff options
context:
space:
mode:
authorGustaf Rydholm <gustaf.rydholm@gmail.com>2022-10-02 01:45:34 +0200
committerGustaf Rydholm <gustaf.rydholm@gmail.com>2022-10-02 01:45:34 +0200
commitffec11ce67d8fe75ea0d5dde5ddf17eb1017fa7d (patch)
treedb8c78232e588b12d7a8b408682783e0b5858615 /text_recognizer/networks/text_decoder.py
parentcf2a827db5798a245dd5207685251675d311dbec (diff)
Add comments
Diffstat (limited to 'text_recognizer/networks/text_decoder.py')
-rw-r--r--text_recognizer/networks/text_decoder.py14
1 files changed, 5 insertions, 9 deletions
diff --git a/text_recognizer/networks/text_decoder.py b/text_recognizer/networks/text_decoder.py
index 7ee6720..7498663 100644
--- a/text_recognizer/networks/text_decoder.py
+++ b/text_recognizer/networks/text_decoder.py
@@ -1,6 +1,4 @@
"""Text decoder."""
-from typing import Optional, Type
-
import torch
from torch import Tensor, nn
@@ -8,26 +6,24 @@ from text_recognizer.networks.transformer.decoder import Decoder
class TextDecoder(nn.Module):
- """Decoder transformer network."""
+ """Decodes images to token logits."""
def __init__(
self,
- hidden_dim: int,
+ dim: int,
num_classes: int,
pad_index: Tensor,
decoder: Decoder,
) -> None:
super().__init__()
- self.hidden_dim = hidden_dim
+ self.dim = dim
self.num_classes = num_classes
self.pad_index = pad_index
self.decoder = decoder
self.token_embedding = nn.Embedding(
- num_embeddings=self.num_classes, embedding_dim=self.hidden_dim
- )
- self.to_logits = nn.Linear(
- in_features=self.hidden_dim, out_features=self.num_classes
+ num_embeddings=self.num_classes, embedding_dim=self.dim
)
+ self.to_logits = nn.Linear(in_features=self.dim, out_features=self.num_classes)
def forward(self, tokens: Tensor, img_features: Tensor) -> Tensor:
"""Decodes latent images embedding into word pieces.