Install with pip install pytorch_pretrained_vit and load a pretrained ViT with:

from pytorch_pretrained_vit import ViT
model = ViT('B_16_imagenet1k', pretrained=True)

Or find a Google Colab example here.

This repository contains an op-for-op PyTorch reimplementation of the Visual Transformer architecture from Google, along with pre-trained models and examples.

The goal of this implementation is to be simple, highly extensible, and easy to integrate into your own projects.

At the moment, you can easily:

• Load pretrained ViT models
• Evaluate on ImageNet or your own data
• Finetune ViT on your own dataset

(Upcoming features) Coming soon:

• Train ViT from scratch on ImageNet (1K)
• Export to ONNX for efficient inference

1. About ViT
2. About ViT-PyTorch
3. Installation
4. Usage
   • Load pretrained models
   • Example: Classify
5. Contributing

Visual Transformers (ViT) are a straightforward application of the transformer architecture to image classification. Even in computer vision, it seems, attention is all you need.

The ViT architecture works as follows: (1) it considers an image as a 1-dimensional sequence of patches, (2) it prepends a classification token to the sequence, (3) it passes these patches through a transformer encoder (like BERT), (4) it passes the first token of the output of the transformer through a small MLP to obtain the classification logits. ViT is trained on a large-scale dataset (ImageNet-21k) with a huge amount of compute.

ViT-PyTorch is a PyTorch re-implementation of ViT. It is consistent with the original Jax implementation, so that it's easy to load Jax-pretrained weights.

At the same time, we aim to make our PyTorch implementation as simple, flexible, and extensible as possible.

Install with pip:

pip install pytorch_pretrained_vit

Or from source:

git clone https://github.com/lukemelas/ViT-PyTorch
cd ViT-Pytorch
pip install -e .

Loading a pretrained model is easy:

from pytorch_pretrained_vit import ViT
model = ViT('B_16_imagenet1k', pretrained=True)

Details about the models are below:

Loading custom configurations is just as easy:

from pytorch_pretrained_vit import ViT
# The following is equivalent to ViT('B_16')
config = dict(hidden_size=512, num_heads=8, num_layers=6)
model = ViT.from_config(config)

Below is a simple, complete example. It may also be found as a Jupyter notebook in examples/simple or as a Colab Notebook.

import json
from PIL import Image
import torch
from torchvision import transforms

# Load ViT
from pytorch_pretrained_vit import ViT
model = ViT('B_16_imagenet1k', pretrained=True)
model.eval()

# Load image
# NOTE: Assumes an image `img.jpg` exists in the current directory
img = transforms.Compose([
    transforms.Resize((384, 384)), 
    transforms.ToTensor(),
    transforms.Normalize(0.5, 0.5),
])(Image.open('img.jpg')).unsqueeze(0)
print(img.shape) # torch.Size([1, 3, 384, 384])

# Classify
with torch.no_grad():
    outputs = model(img)
print(outputs.shape)  # (1, 1000)

See examples/imagenet for details about evaluating on ImageNet.

Other great repositories with this model include:

• Ross Wightman's repo
• Phil Wang's repo

If you find a bug, create a GitHub issue, or even better, submit a pull request. Similarly, if you have questions, simply post them as GitHub issues.

I look forward to seeing what the community does with these models!