CLIP: Learning Transferable Models for Multimodal Learning

Authors: Radford et al. Year: 2021 ArXiv/Link: https://arxiv.org/abs/2103.14030

Summary

Learns visual representations by aligning images with text descriptions through contrastive learning at scale.

Key Concepts

• Vision-language alignment
• Contrastive learning
• Zero-shot learning
• Transferability
• Multimodal embeddings

Impact

Enabled zero-shot vision classification and multimodal applications

Category

Vision-Language Models

---