We develop a variational approach to solve image inverse problems like super resolution, inpainting and deblurring for flow-based generative models.

OnlineSplatter enables pose-free 3D reconstruction of free-moving objects from monocular video by fusing frame-wise features into a compact Gaussian field via a dual-key memory, with constant-time and memory efficiency.

We propose KRIS-Bench, a benchmark designed to evaluate knowledge-based reasoning in instruction-based image editing models.

This papers introduces a Diffusion Transformer (DiT) for sampling 3D molecular conformers.

We provide a faster algorithm for generic identification in tree-shaped linear structural causal models.

A new model with inherent mechanistic and concept-based explanations and a new concept consistency evaluation framework