Latent Spatial Memory stores persistent 3D scene content directly as latent tokens for efficient, spatially consistent video world models.

TriSplat predicts simulation-ready triangle primitives for feed-forward sparse-view 3D scene reconstruction.

World-R1 aligns text-to-video generation with 3D constraints through reinforcement learning, improving geometric consistency while preserving visual quality and motion diversity.

We present a problem-driven survey of feed-forward 3D scene modeling, covering representative methods, datasets, and downstream applications.

We propose Chain-of-View (CoV) prompting, a training-free test-time reasoning framework that transforms VLMs into active viewpoint reasoners for spatial reasoning in 3D environments.

We present DriveGen3D, a novel framework for generating high-quality and highly controllable dynamic 3D driving scenes that addresses critical limitations in existing methodologies.

VolSplat improves multi-view consistency and geometric accuracy for feed-forward 3DGS with voxel-aligned prediction.

We introduce PM-Loss, a novel regularization loss based on a learned point map for feed-forward 3DGS, leading to smoother 3D geometry and better rendering.

ZPressor is an architecture-agnostic module that compresses multi-view inputs for scalable feed-forward 3DGS.

We introduce WonderTurbo, the first real-time interactive 3D scene generation framework capable of generating novel perspectives of 3D scenes within 0.72 seconds.

We introduce Flash-GRPO, a one-step policy optimization framework that improves video diffusion alignment efficiency with iso-temporal grouping and temporal gradient rectification.

We introduce DiGSeg, a diffusion-based generalist segmentation framework that repurposes pretrained diffusion models for text-conditioned semantic and open-vocabulary segmentation.

We introduce MV-RoboBench, a benchmark for evaluating multi-view spatial reasoning in robotic manipulation, revealing large gaps between state-of-the-art VLMs and human performance.

We introduce ReVisual-R1, achieving a new state-of-the-art among open-source 7B MLLMs on challenging benchmarks.

We propose a single-view RGB-D-based depth completion framework, TransDiff, that leverages the Denoising Diffusion Probabilistic Models(DDPM) to achieve material-agnostic object grasping in desktop.

This paper aims to tackle the problem of modeling dynamic urban street scenes from monocular videos. We introduce Street Gaussians, a new explicit scene representation that tackles some major limitations.