Learning Transferable Visual Models From Natural Language Supervision

TL;DR

This paper introduces a method for training visual models using natural language supervision from large-scale image-caption pairs, enabling zero-shot transfer to various vision tasks without task-specific training.

Contribution

The authors demonstrate that simple image-caption matching pre-training on 400 million pairs yields state-of-the-art representations capable of zero-shot transfer across diverse tasks.

Findings

Achieves competitive zero-shot performance on 30+ datasets.

Matches ImageNet accuracy without using ImageNet training data.

Enables flexible referencing of visual concepts through natural language.

Abstract

State-of-the-art computer vision systems are trained to predict a fixed set of predetermined object categories. This restricted form of supervision limits their generality and usability since additional labeled data is needed to specify any other visual concept. Learning directly from raw text about images is a promising alternative which leverages a much broader source of supervision. We demonstrate that the simple pre-training task of predicting which caption goes with which image is an efficient and scalable way to learn SOTA image representations from scratch on a dataset of 400 million (image, text) pairs collected from the internet. After pre-training, natural language is used to reference learned visual concepts (or describe new ones) enabling zero-shot transfer of the model to downstream tasks. We study the performance of this approach by benchmarking on over 30 different existing computer vision datasets, spanning tasks such as OCR, action recognition in videos, geo-localization, and many types of fine-grained object classification. The model transfers non-trivially to most tasks and is often competitive with a fully supervised baseline without the need for any dataset specific training. For instance, we match the accuracy of the original ResNet-50 on ImageNet zero-shot without needing to use any of the 1.28 million training examples it was trained on. We release our code and pre-trained model weights at https://github.com/OpenAI/CLIP.