Daniel Cremers

NOVA3R: Non-pixel-aligned Visual Transformer for Amodal 3D Reconstruction

FlowFeat: Pixel-Dense Embedding of Motion Profiles

Implicit Neural Surface Deformation with Explicit Velocity Fields

True Multimodal In-Context Learning Needs Attention to the Visual Context

IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals

Laplace-Transform-Filters render spectral Graph Neural Networks transferable

Uncertainty-Based Abstention in LLMs Improves Safety and Reduces Hallucinations