ImageBind One Embedding Space to Bind Them All

🤖 Plain-English Summary

We present ImageBind, an approach to learn a joint embedding across six different modalities - images, text, audio, depth, thermal, and IMU data. The emergent capabilities improve with the strength of the image encoder and we set a new advanced on emergent zero-shot recognition tasks across modalities, outperforming specialist supervised models.

🔑 Key Findings

• We show that all combinations of paired data are not necessary to train such a joint embedding, and only image-paired data is sufficient to bind the modalities together.
• ImageBind can leverage recent large scale vision-language models, and extends their zero-shot capabilities to new modalities just by using their natural pairing with images.
• It enables novel emergent applications ‘out-of-the-box’ including cross-modal retrieval, composing modalities with arithmetic, cross-modal detection and generation.

🔗 Related Resources