While standard image models train an image feature extractor and a linear classifier to predict the probability distribution over class labels, CLIP jointly trains an image encoder and a text encoder to predict the correct pairings of a batch of (image, text) training examples. At test time the learned text encoder synthesizes a zero-shot linear classifier by embedding the names or descriptions of the target dataset’s classes.
CLIPForImageClassification Model Card, OpenAI CLIP Model Card
Our CLIPForImageClassification model on hub
Google ViTForImageClassification model on hub
Our ViTForImageClassification model on hub
@misc{radford2021learning,
title={Learning Transferable Visual Models From Natural Language Supervision},
author={Alec Radford and Jong Wook Kim and Chris Hallacy and Aditya Ramesh and Gabriel Goh and Sandhini Agarwal and Girish Sastry and Amanda Askell and Pamela Mishkin and Jack Clark and Gretchen Krueger and Ilya Sutskever},
year={2021},
eprint={2103.00020},
archivePrefix={arXiv},
primaryClass={cs.CV}
}@misc{dosovitskiy2021image,
title={An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale},
author={Alexey Dosovitskiy and Lucas Beyer and Alexander Kolesnikov and Dirk Weissenborn and Xiaohua Zhai and Thomas Unterthiner and Mostafa Dehghani and Matthias Minderer and Georg Heigold and Sylvain Gelly and Jakob Uszkoreit and Neil Houlsby},
year={2021},
eprint={2010.11929},
archivePrefix={arXiv},
primaryClass={cs.CV}
}