D4RT

Researchers introduced GenCeption, a model that uses a pre-trained video generative diffusion backbone as a general-purpose vision learner, achieving state-of-the-art performance on tasks like depth e…