#

Vision Transformer (ViT)

This is a PyTorch implementation of the paper An Image Is Worth 16x16 Words: Transformers For Image Recognition At Scale.

Vision transformer applies a pure transformer to images without any convolution layers. They split the image into patches and apply a transformer on patch embeddings. Patch embeddings are generated by applying a simple linear transformation to the flattened pixel values of the patch. Then a standard transformer encoder is fed with the patch embeddings, along with a classification token [CLS] . The encoding on the [CLS] token is used to classify the image with an MLP.

When feeding the transformer with the patches, learned positional embeddings are added to the patch embeddings, because the patch embeddings do not have any information about where that patch is from. The positional embeddings are a set of vectors for each patch location that get trained with gradient descent along with other parameters.

ViTs perform well when they are pre-trained on large datasets. The paper suggests pre-training them with an MLP classification head and then using a single linear layer when fine-tuning. The paper beats SOTA with a ViT pre-trained on a 300 million image dataset. They also use higher resolution images during inference while keeping the patch size the same. The positional embeddings for new patch locations are calculated by interpolating learning positional embeddings.

Here's an experiment that trains ViT on CIFAR-10. This doesn't do very well because it's trained on a small dataset. It's a simple experiment that anyone can run and play with ViTs.

43import torch
44from torch import nn
45
46from labml_helpers.module import Module
47from labml_nn.transformers import TransformerLayer
48from labml_nn.utils import clone_module_list

#

Get patch embeddings

The paper splits the image into patches of equal size and do a linear transformation on the flattened pixels for each patch.

We implement the same thing through a convolution layer, because it's simpler to implement.

51class PatchEmbeddings(Module):

#

d_model is the transformer embeddings size
patch_size is the size of the patch
in_channels is the number of channels in the input image (3 for rgb)

63    def __init__(self, d_model: int, patch_size: int, in_channels: int):

#

69        super().__init__()

#

We create a convolution layer with a kernel size and and stride length equal to patch size. This is equivalent to splitting the image into patches and doing a linear transformation on each patch.

74        self.conv = nn.Conv2d(in_channels, d_model, patch_size, stride=patch_size)

#

x is the input image of shape [batch_size, channels, height, width]

76    def forward(self, x: torch.Tensor):

#

Apply convolution layer

81        x = self.conv(x)

#

Get the shape.

83        bs, c, h, w = x.shape

#

Rearrange to shape [patches, batch_size, d_model]

85        x = x.permute(2, 3, 0, 1)
86        x = x.view(h * w, bs, c)

#

Return the patch embeddings

89        return x

#

Add parameterized positional encodings

This adds learned positional embeddings to patch embeddings.

92class LearnedPositionalEmbeddings(Module):

#

d_model is the transformer embeddings size
max_len is the maximum number of patches

101    def __init__(self, d_model: int, max_len: int = 5_000):

#

106        super().__init__()

#

Positional embeddings for each location

108        self.positional_encodings = nn.Parameter(torch.zeros(max_len, 1, d_model), requires_grad=True)

#

x is the patch embeddings of shape [patches, batch_size, d_model]

110    def forward(self, x: torch.Tensor):

#

Get the positional embeddings for the given patches

115        pe = self.positional_encodings[:x.shape[0]]

#

Add to patch embeddings and return

117        return x + pe

#

MLP Classification Head

This is the two layer MLP head to classify the image based on [CLS] token embedding.

120class ClassificationHead(Module):

#

d_model is the transformer embedding size
n_hidden is the size of the hidden layer
n_classes is the number of classes in the classification task

128    def __init__(self, d_model: int, n_hidden: int, n_classes: int):

#

134        super().__init__()

#

First layer

136        self.linear1 = nn.Linear(d_model, n_hidden)

#

Activation

138        self.act = nn.ReLU()

#

Second layer

140        self.linear2 = nn.Linear(n_hidden, n_classes)

#

x is the transformer encoding for [CLS] token

142    def forward(self, x: torch.Tensor):

#

First layer and activation

147        x = self.act(self.linear1(x))

#

Second layer

149        x = self.linear2(x)

#

152        return x

#

Vision Transformer

This combines the patch embeddings, positional embeddings, transformer and the classification head.

155class VisionTransformer(Module):

#

transformer_layer is a copy of a single transformer layer. We make copies of it to make the transformer with n_layers .
n_layers is the number of transformer layers.
patch_emb is the patch embeddings layer.
pos_emb is the positional embeddings layer.
classification is the classification head.

163    def __init__(self, transformer_layer: TransformerLayer, n_layers: int,
164                 patch_emb: PatchEmbeddings, pos_emb: LearnedPositionalEmbeddings,
165                 classification: ClassificationHead):

#

174        super().__init__()

#

Patch embeddings

176        self.patch_emb = patch_emb
177        self.pos_emb = pos_emb

#

Classification head

179        self.classification = classification

#

Make copies of the transformer layer

181        self.transformer_layers = clone_module_list(transformer_layer, n_layers)

#

[CLS] token embedding

184        self.cls_token_emb = nn.Parameter(torch.randn(1, 1, transformer_layer.size), requires_grad=True)

#

Final normalization layer

186        self.ln = nn.LayerNorm([transformer_layer.size])

#

x is the input image of shape [batch_size, channels, height, width]

188    def forward(self, x: torch.Tensor):

#

Get patch embeddings. This gives a tensor of shape [patches, batch_size, d_model]

193        x = self.patch_emb(x)

#

Concatenate the [CLS] token embeddings before feeding the transformer

195        cls_token_emb = self.cls_token_emb.expand(-1, x.shape[1], -1)
196        x = torch.cat([cls_token_emb, x])

#

Add positional embeddings

198        x = self.pos_emb(x)

#

Pass through transformer layers with no attention masking

201        for layer in self.transformer_layers:
202            x = layer(x=x, mask=None)

#

Get the transformer output of the [CLS] token (which is the first in the sequence).

205        x = x[0]

#

Layer normalization

208        x = self.ln(x)

#

Classification head, to get logits

211        x = self.classification(x)

#

214        return x