Publications

Déjà View: Looping Transformers for Multi-View 3D Reconstruction

Déjà View applies a single looped transformer block recurrently to per-view features for K refinement steps, making explicit the iterative refinement that multi-view reconstruction transformers otherwise buy inefficiently through unique parameters …

ArtiFixer: Enhancing and Extending 3D Reconstruction with Auto-Regressive Diffusion Models

Per-scene optimization methods such as 3D Gaussian Splatting provide state-of-the-art novel view synthesis quality but extrapolate poorly to under-observed regions. ArtiFixer bridges 3D neural reconstruction and auto-regressive video generation to …

MapAnything: Universal Feed-Forward Metric 3D Reconstruction

MapAnything is a unified transformer-based feed-forward model that ingests one or more images along with optional geometric inputs such as camera intrinsics, poses, depth, or partial reconstructions, and directly regresses metric 3D scene geometry …

FlowR: Flowing from Sparse to Dense 3D Reconstructions

We present FlowR, a novel pipeline that bridges the gap between sparse and dense 3D reconstruction. Contrary to prior works, we learn a direct mapping between incorrect renderings and their corresponding ground-truth images, augmenting scene captures with consistent novel, generated views to improve reconstruction quality.

Dynamic 3D Gaussian Fields for Urban Areas

Given a set of *heterogeneous* input sequences of a common geographic area, we optimize a *single* dynamic scene representation that permits rendering of arbitrary viewpoints and scene configurations at interactive speeds.

Multi-Level Neural Scene Graphs for Dynamic Urban Environments

We present the *first benchmark* and a *novel method* for radiance field reconstruction of dynamic urban areas from *heterogeneous, multi-sequence data*.

R3D3: Dense 3D Reconstruction of Dynamic Scenes from Multiple Cameras

We present R3D3, a method for dense 3D reconstruction and ego-motion estimation in dynamic outdoor environments that iterates between geometric multi-camera optimization and monocular depth refinement.

Monocular Quasi-Dense 3D Object Tracking

A reliable and accurate 3D tracking framework is essential for predicting future locations of surrounding objects and planning the observer's actions in numerous applications such as autonomous driving. We propose a framework that can effectively …

Track to Reconstruct and Reconstruct to Track

Object tracking and 3D reconstruction are often performed together, with tracking used as input for reconstruction. However, the obtained reconstructions also provide useful information for improving tracking. We propose a novel method that closes …