ImageBind

Multimodal is becoming the norm, binding data from six modalities at once without explicit supervision is impressive. The relationships it recognizes between images, video, audio, text, depth, thermal and IMUs could be a game-changer. Looking forward to testing it out.