Omprakash Chakraborty

Standard image ViTs, when adapted with masked Siamese learning on "super images," unlock state-of-the-art video representation learning with orders of magnitude less pretraining.

State-of-the-art semi-supervised domain generalization (SSDG) methods crumble when faced with the real-world challenge of long-tailed class distributions, but IMaX offers a simple, effective fix.

Forget prompt engineering – ORION unlocks surprisingly large gains in VLM performance by simply fine-tuning text embeddings to be more orthogonal.