FAR (feature-to-action regression) is a simple and fast WAM recipe with highly competitive performance. It predicts actions directly from frozen visual features, without future imagination or iterative denoising.
LDR learns "how the future evolves" rather than "what the future is". It's one of the first video world models that capture the underlying dynamics purely from pixels and extrapolate them beyond the training distribution.
Powered by large-scale curated data and a sphere-aware ViT, DA2 predicts dense distance from a single 360° panorama in an end-to-end manner, with remarkable geometric fidelity and strong zero-shot generalization.
Based on Stable Diffusion, Lotus delivers SOTA performance on monocular depth & normal estimation with a simple yet effective fine-tuning protocol that better fits the pre-trained visual prior for dense prediction.