Jing Liao

F2DR evaluates DeepSearch workflows across three dimensions: Content, Trajectory, and Answer, enabling comprehensive process-level assessment, and constructs DeepSearch RM-Bench, a dedicated benchmark for evaluating RMs in DeepSearch scenarios.

ThinkBLOX transforms 3D scene generation by enabling iterative refinement, resulting in layouts that are both physically plausible and semantically coherent.

BoxCtrl's innovative use of RGB 3D bounding boxes enables unprecedented precision in geometric image editing, outperforming traditional methods.

CanonicalGS achieves a remarkable 2.5 dB improvement in novel view synthesis quality by stabilizing scene representations against noisy input data.

Achieving state-of-the-art mesh quality with 18x faster inference times by directly generating triangle soups while respecting crucial symmetries.

VLMs can outperform traditional problem-solving roles by serving as adaptive teachers, leading to a dramatic 16.7-point boost in video reasoning performance.

Skip the costly VFX pipeline: EffectMaker lets you create customized visual effects with a reference video and a multimodal LLM, no per-effect fine-tuning needed.

Achieve photorealistic 3D human animation by guiding video diffusion with kinematic renderings and a novel self-guided stochastic sampling method that overcomes out-of-distribution challenges.