Learning Transferable Visual Models From Natural Language Supervision

TLDR

Current computer vision models are trained on a fixed set of object categories, limiting generality because new concepts require additional labeled data. The study proposes learning visual representations directly from raw image‑text pairs, leveraging broad natural language supervision. The authors pre‑train a model to match images with captions on 400 million internet image‑text pairs, then use language to refer to learned visual concepts, enabling zero‑shot transfer to diverse vision tasks. The resulting model achieves competitive zero‑shot performance on most tasks, matching ResNet‑50 ImageNet accuracy without training data, and the code and weights are publicly released.

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.