Vit Large Patch16 224 In21k

🚀 Vision Transformer (large-sized model)

A Vision Transformer (ViT) model pre-trained on ImageNet-21k, suitable for extracting image features and fine-tuning on downstream tasks.

🚀 Quick Start

The Vision Transformer (ViT) is a transformer encoder model (similar to BERT) pre-trained on a large collection of images in a supervised manner, specifically ImageNet-21k, at a resolution of 224x224 pixels.

# Code examples remain unchanged
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-large-patch16-224-in21k')
model = ViTModel.from_pretrained('google/vit-large-patch16-224-in21k')

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

✨ Features

• Pretrained on Large-scale Dataset: Trained on ImageNet-21k, which contains 14 million images and 21,843 classes, enabling it to learn rich image representations.
• Flexible for Downstream Tasks: The model includes a pre-trained pooler, which can be used for downstream tasks such as image classification.
• Standard Transformer Architecture: Images are presented as a sequence of fixed-size patches, and absolute position embeddings are added, following the standard Transformer architecture.

📚 Documentation

Model description

The Vision Transformer (ViT) is a transformer encoder model (BERT-like) pretrained on a large collection of images in a supervised fashion, namely ImageNet-21k, at a resolution of 224x224 pixels.

Images are presented to the model as a sequence of fixed-size patches (resolution 16x16), which are linearly embedded. One also adds a [CLS] token to the beginning of a sequence to use it for classification tasks. One also adds absolute position embeddings before feeding the sequence to the layers of the Transformer encoder.

Note that this model does not provide any fine-tuned heads, as these were zero'd by Google researchers. However, the model does include 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: if you have a dataset of labeled images for instance, you can train a standard classifier by placing a linear layer on top of the pre-trained encoder. One typically places a linear layer 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 to embed images, but it's mostly intended to be fine-tuned on a downstream task.

How to use

Here is how to use this model:

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-large-patch16-224-in21k')
model = ViTModel.from_pretrained('google/vit-large-patch16-224-in21k')

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

Currently, both the feature extractor and model support PyTorch. Tensorflow and JAX/FLAX are coming soon, and the API of ViTFeatureExtractor might change.

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.

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}
}

📄 License

This project is licensed under the Apache-2.0 license.