summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorGustaf Rydholm <gustaf.rydholm@gmail.com>2021-09-18 17:43:44 +0200
committerGustaf Rydholm <gustaf.rydholm@gmail.com>2021-09-18 17:43:44 +0200
commit1a193f09bd9199609c30a005dbd28f587dce2606 (patch)
treed548fcc9c22b00e23512473b1cf8ceddc5121716
parentab47938bfa2d70e6244e8431707586e32f2c7d50 (diff)
Add Vq transfomer model
-rw-r--r--text_recognizer/models/vq_transformer.py99
-rw-r--r--text_recognizer/networks/vq_transformer.py67
2 files changed, 154 insertions, 12 deletions
diff --git a/text_recognizer/models/vq_transformer.py b/text_recognizer/models/vq_transformer.py
new file mode 100644
index 0000000..71ca2ef
--- /dev/null
+++ b/text_recognizer/models/vq_transformer.py
@@ -0,0 +1,99 @@
+"""PyTorch Lightning model for base Transformers."""
+from typing import Tuple, Type, Set
+
+import attr
+import torch
+from torch import Tensor
+
+from text_recognizer.models.metrics import CharacterErrorRate
+from text_recognizer.models.transformer import TransformerLitModel
+
+
+@attr.s(auto_attribs=True, eq=False)
+class VqTransformerLitModel(TransformerLitModel):
+ """A PyTorch Lightning model for transformer networks."""
+
+ def forward(self, data: Tensor) -> Tensor:
+ """Forward pass with the transformer network."""
+ return self.predict(data)
+
+ def training_step(self, batch: Tuple[Tensor, Tensor], batch_idx: int) -> Tensor:
+ """Training step."""
+ data, targets = batch
+ logits, commitment_loss = self.network(data, targets[:-1])
+ loss = self.loss_fn(logits, targets[1:]) + commitment_loss
+ self.log("train/loss", loss)
+ self.log("train/commitment_loss", commitment_loss)
+ return loss
+
+ def validation_step(self, batch: Tuple[Tensor, Tensor], batch_idx: int) -> None:
+ """Validation step."""
+ data, targets = batch
+
+ # Compute the loss.
+ logits, commitment_loss = self.network(data, targets[:-1])
+ loss = self.loss_fn(logits, targets[1:]) + commitment_loss
+ self.log("val/loss", loss, prog_bar=True)
+ self.log("val/commitment_loss", commitment_loss)
+
+ # Get the token prediction.
+ pred = self(data)
+ self.val_cer(pred, targets)
+ self.log("val/cer", self.val_cer, on_step=False, on_epoch=True, prog_bar=True)
+ self.test_acc(pred, targets)
+ self.log("val/acc", self.test_acc, on_step=False, on_epoch=True)
+
+ def test_step(self, batch: Tuple[Tensor, Tensor], batch_idx: int) -> None:
+ """Test step."""
+ data, targets = batch
+
+ # Compute the text prediction.
+ pred = self(data)
+ self.test_cer(pred, targets)
+ self.log("test/cer", self.test_cer, on_step=False, on_epoch=True, prog_bar=True)
+ self.test_acc(pred, targets)
+ self.log("test/acc", self.test_acc, on_step=False, on_epoch=True)
+
+ def predict(self, x: Tensor) -> Tensor:
+ """Predicts text in image.
+
+ Args:
+ x (Tensor): Image(s) to extract text from.
+
+ Shapes:
+ - x: :math: `(B, H, W)`
+ - output: :math: `(B, S)`
+
+ Returns:
+ Tensor: A tensor of token indices of the predictions from the model.
+ """
+ bsz = x.shape[0]
+
+ # Encode image(s) to latent vectors.
+ z, _ = self.network.encode(x)
+
+ # Create a placeholder matrix for storing outputs from the network
+ output = torch.ones((bsz, self.max_output_len), dtype=torch.long).to(x.device)
+ output[:, 0] = self.start_index
+
+ for Sy in range(1, self.max_output_len):
+ context = output[:, :Sy] # (B, Sy)
+ logits = self.network.decode(z, context) # (B, Sy, C)
+ tokens = torch.argmax(logits, dim=-1) # (B, Sy)
+ output[:, Sy : Sy + 1] = tokens[:, -1:]
+
+ # Early stopping of prediction loop if token is end or padding token.
+ if (
+ (output[:, Sy - 1] == self.end_index)
+ | (output[:, Sy - 1] == self.pad_index)
+ ).all():
+ break
+
+ # Set all tokens after end token to pad token.
+ for Sy in range(1, self.max_output_len):
+ idx = (output[:, Sy - 1] == self.end_index) | (
+ output[:, Sy - 1] == self.pad_index
+ )
+ output[idx, Sy] = self.pad_index
+
+ return output
diff --git a/text_recognizer/networks/vq_transformer.py b/text_recognizer/networks/vq_transformer.py
index 0433863..69f68fd 100644
--- a/text_recognizer/networks/vq_transformer.py
+++ b/text_recognizer/networks/vq_transformer.py
@@ -1,5 +1,6 @@
"""Vector quantized encoder, transformer decoder."""
-from typing import Tuple
+from pathlib import Path
+from typing import Tuple, Optional
import torch
from torch import Tensor
@@ -22,7 +23,8 @@ class VqTransformer(ConvTransformer):
pad_index: Tensor,
encoder: VQVAE,
decoder: Decoder,
- pretrained_encoder_path: str,
+ no_grad: bool,
+ pretrained_encoder_path: Optional[str] = None,
) -> None:
super().__init__(
input_dims=input_dims,
@@ -34,18 +36,36 @@ class VqTransformer(ConvTransformer):
encoder=encoder,
decoder=decoder,
)
- self.pretrained_encoder_path = pretrained_encoder_path
-
# For typing
self.encoder: VQVAE
- def setup_encoder(self) -> None:
+ self.no_grad = no_grad
+
+ if pretrained_encoder_path is not None:
+ self.pretrained_encoder_path = (
+ Path(__file__).resolve().parents[2] / pretrained_encoder_path
+ )
+ self._setup_encoder()
+ else:
+ self.pretrained_encoder_path = None
+
+ def _load_pretrained_encoder(self) -> None:
+ self.encoder.load_state_dict(
+ torch.load(self.pretrained_encoder_path)["state_dict"]["network"]
+ )
+
+ def _setup_encoder(self) -> None:
"""Remove unecessary layers."""
- # TODO: load pretrained vqvae
+ self._load_pretrained_encoder()
del self.encoder.decoder
- del self.encoder.post_codebook_conv
+ # del self.encoder.post_codebook_conv
+
+ def _encode(self, x: Tensor) -> Tuple[Tensor, Tensor]:
+ z_e = self.encoder.encode(x)
+ z_q, commitment_loss = self.encoder.quantize(z_e)
+ return z_q, commitment_loss
- def encode(self, x: Tensor) -> Tensor:
+ def encode(self, x: Tensor) -> Tuple[Tensor, Tensor]:
"""Encodes an image into a discrete (VQ) latent representation.
Args:
@@ -62,12 +82,35 @@ class VqTransformer(ConvTransformer):
Returns:
Tensor: A Latent embedding of the image.
"""
- with torch.no_grad():
- z_e = self.encoder.encode(x)
- z_q, _ = self.encoder.quantize(z_e)
+ if self.no_grad:
+ with torch.no_grad():
+ z_q, commitment_loss = self._encode(x)
+ else:
+ z_q, commitment_loss = self._encode(x)
z = self.latent_encoder(z_q)
# Permute tensor from [B, E, Ho * Wo] to [B, Sx, E]
z = z.permute(0, 2, 1)
- return z
+ return z, commitment_loss
+
+ def forward(self, x: Tensor, context: Tensor) -> Tensor:
+ """Encodes images into word piece logtis.
+
+ Args:
+ x (Tensor): Input image(s).
+ context (Tensor): Target word embeddings.
+
+ Shapes:
+ - x: :math: `(B, C, H, W)`
+ - context: :math: `(B, Sy, T)`
+
+ where B is the batch size, C is the number of input channels, H is
+ the image height and W is the image width.
+
+ Returns:
+ Tensor: Sequence of logits.
+ """
+ z, commitment_loss = self.encode(x)
+ logits = self.decode(z, context)
+ return logits, commitment_loss