Vit Base Patch32 224 In21k

🚀 Vision Transformer (base-sized model)

A Vision Transformer (ViT) model pre-trained on ImageNet-21k, offering a powerful solution for image recognition tasks.

🚀 Quick Start

The Vision Transformer (ViT) is a transformer encoder model (similar to BERT) that has been pre-trained on a large collection of images, specifically ImageNet-21k, at a resolution of 224x224 pixels. It was introduced in the paper An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale by Dosovitskiy et al. and first released in this repository.

✨ Features

• Large-scale Pre-training: Trained on ImageNet-21k with 14 million images and 21,843 classes.
• Transformer-based Architecture: Utilizes a BERT-like encoder for image processing.
• Flexible for Downstream Tasks: Can be used for various image-related tasks with a linear layer on top.

📦 Installation

No specific installation steps are provided in the original document.

💻 Usage Examples

Basic Usage

from transformers import ViTImageProcessor, ViTModel
from PIL import Image
import requests

url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
image = Image.open(requests.get(url, stream=True).raw)

processor = ViTImageProcessor.from_pretrained('google/vit-base-patch32-224-in21k')
model = ViTModel.from_pretrained('google/vit-base-patch32-224-in21k')

inputs = processor(images=image, return_tensors="pt")
outputs = model(**inputs)
last_hidden_state = outputs.last_hidden_state

Refer to the docs for usage in TensorFlow and JAX/FLAX.

📚 Documentation

Model description

Images are presented to the model as a sequence of fixed-size patches (resolution 32x32), which are linearly embedded. A [CLS] token is added to the beginning of a sequence for classification tasks, and absolute position embeddings are added before feeding the sequence to the Transformer encoder layers.

Note that this model does not provide any fine-tuned heads, as these were zero'd by Google researchers. However, it includes the pre-trained pooler, which can be used for downstream tasks (such as image classification).

By pre-training the model, it learns an inner representation of images that can then be used to extract features useful for downstream tasks. For example, you can train a standard classifier by placing a linear layer on top of the pre-trained encoder. Typically, a linear layer is placed on top of the [CLS] token, as the last hidden state of this token can be seen as a representation of an entire image.

Intended uses & limitations

You can use the raw model for image classification. See the model hub to look for fine-tuned versions on a task that interests you.

Training data

The ViT model was pretrained on ImageNet-21k, a dataset consisting of 14 million images and 21k classes.

Training procedure

Preprocessing

The exact details of preprocessing of images during training/validation can be found here. Images are resized/rescaled to the same resolution (224x224) and normalized across the RGB channels with mean (0.5, 0.5, 0.5) and standard deviation (0.5, 0.5, 0.5).

Pretraining

The model was trained on TPUv3 hardware (8 cores). All model variants are trained with a batch size of 4096 and learning rate warmup of 10k steps. For ImageNet, the authors found it beneficial to additionally apply gradient clipping at global norm 1. Pre-training resolution is 224.

Evaluation results

For evaluation results on several image classification benchmarks, we refer to tables 2 and 5 of the original paper. Note that for fine-tuning, the best results are obtained with a higher resolution (384x384). Of course, increasing the model size will result in better performance.

🔧 Technical Details

The model architecture is based on a transformer encoder, similar to BERT. It processes images by converting them into a sequence of patches and then applying linear embeddings, position embeddings, and the Transformer encoder layers. The pre-training on ImageNet-21k allows the model to learn general image features that can be useful for various downstream tasks.

📄 License

This project is licensed under the Apache-2.0 license.

BibTeX entry and citation info

@misc{wu2020visual,
      title={Visual Transformers: Token-based Image Representation and Processing for Computer Vision}, 
      author={Bichen Wu and Chenfeng Xu and Xiaoliang Dai and Alvin Wan and Peizhao Zhang and Zhicheng Yan and Masayoshi Tomizuka and Joseph Gonzalez and Kurt Keutzer and Peter Vajda},
      year={2020},
      eprint={2006.03677},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

@inproceedings{deng2009imagenet,
  title={Imagenet: A large-scale hierarchical image database},
  author={Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Li, Kai and Fei-Fei, Li},
  booktitle={2009 IEEE conference on computer vision and pattern recognition},
  pages={248--255},
  year={2009},
  organization={Ieee}
}