First explored vLLM-based agentic interaction in generated 3D contents.

First benchmarking framework for evaluating vision-language model scene understanding via inverse rendering tasks through agentic tool use.

JarvisArt outperforms GPT-4o with a 60% improvement in average pixel-level metrics on MMArt-Bench for content fidelity.

DynamicVerse is a physical‑scale, multimodal 4D modeling framework for real-world video.

Enhance multimodal fusion via conditional mutual information maximization for balanced cross-modal representation learning.

A granular video anomaly detection framework that integrates the detection of multiple fine-grained anomalous objects into a unified framework, achieving state-of-the-art performance.

Train once, deploy everywhere with many-in-one flexible 3D Gaussian splatting for efficient multi-device deployment.

JarvisIR is a VLM-powered intelligent system that dynamically schedules expert models for restoration.

An initial exploration into embedding customizable, imperceptible, and recoverable information within the renders produced by off-the-line 3D generative models, while ensuring minimal impact on the rendered content's quality.

A pioneering foray into the intriguing realm of embedding, relating and perceiving the heterogeneous patterns from various biomedical modalities holistically via a graph theory.

A pioneering exploration into high-fidelity medical video generation on endoscopy scenes.

A novel approach that leverages the ambiguity and uncertainty in object boundaries to improve segmentation performance, turning traditional segmentation "flaws" into advantages.

A probabilistic prompting framework that enhances SAM's performance on ambiguous medical images through uncertainty-aware prompt generation.

Embedding multimodal invisible information (image, video, audio) into distributed visual assets.

Training large networks efficiently and smartly by progressive data distillation.