💫 You-Zhe Xie

YoCausal: How Far is Video Generation from World Model? A Causality Perspective

You-Zhe Xie*, Yu-Hsuan Li*, Jie-Ying Lee, Kaipeng Zhang, Yu-Lun Liu, Zhixiang Wang

Authorship: 1st author

YoCausal is the first benchmark evaluating causal cognition in video generation models, inspired by cognitive science experiments that test whether infants perceive causality using reversed videos. Our benchmark can incorporate any real-world video at zero cost, making it arbitrarily extensible to easily assess video generation models' understanding of diverse types of causality.

  project page   arXiv   Code   Dataset   Video 🤗  Huggingface

Every Camera Effect, Every Time, All at Once: 4D Gaussian Ray Tracing for Physics-based Camera Effect Data Generation

Yi-Ruei Liu*, You-Zhe Xie*, Yu-Hsiang Hsu*, I-Sheng Fang, Yu-Lun Liu, Jun-Cheng Chen

Authorship: co-1st author

✅NeurIPS 2025 SpaVLE Workshop (✨Oral)

4DGRT identify problems about limited understanding of camera-effect video generation model, and proposes a data generation pipeline to synthesize videos with physically-accurate camera effect through apply ray tracing on dynamic 3DGS (4DGS). 4DGRT also construct and release a paired dataset containing 8 dynamic indoor scenes with multi-view videos under four camera effects.

  project page   arXiv   Code   Video

StealthAttack: Robust 3D Gaussian Splatting Poisoning via Density-Guided Illusions

Bo-Hsu Ke, You-Zhe Xie, Yu-Lun Liu, Wei-Chen Chiu

Authorship: 2nd author

✅ICCV 2025 (Poster)

✅ICCV 2025 Responsible Imaging Workshop (✨ Oral, 🎊Best Poster Award)

StealthAttack proposes a novel density-guided poisoning method for 3D Gaussian Splatting (3DGS) that strategically injects illusory objects into low-density regions to create viewpoint-dependent visual illusions. The method combines point cloud poisoning with adaptive noise scheduling to disrupt multi-view consistency, enabling successful attacks where illusions are clearly visible from target views while maintaining high fidelity in innocent viewpoints.

  project page   arXiv   Code   Video 🤗  Huggingface

ShellBench: A Parametric Shell Game Benchmark for Memory-Aware Manipulation

06/2026

ShellBench is a parametric Isaac Sim benchmark that evaluates the long-term memory capacity of policies via a shell game: a ball hidden under shuffled identical cups forces reliance on memory over current observations, with configurable cups, shuffles, and speed yielding quantitative memory-capacity curves.

  Project Page   Code

Designing Fun: LLM-based Game Level Generation via Agent Gameplay Feedback

06/2026

We train an LLM to generate fun MiniGrid levels via a two-stage SFT + GRPO framework. We decompose "fun" into three gameplay-grounded rewards—regret (skill discrimination), interaction richness, and diversity—optimized through agent playthroughs. Our method beats vanilla and SFT baselines on all metrics, with ablations confirming each reward's distinct contribution.

  Project Page   Code

MATCHA: Match with Agent

04/2026

MATCHA is a multi-agent career resource matching system built for the 2026 YTP Hackathon. Its core idea is to let AI Agents socialize on behalf of users: each personal Agent learns the user's background, talks with resource Agents and other user Agents, and brings back useful opportunities or peer experiences. By combining User Agents, government resource Agents, RAG-based policy matching, Coffee Chat discovery, and a government dashboard, MATCHA turns passive searching into active matching.

  Intro Page   Code

Desktop Organizer

12/2025

An application (web demo, React-based) designed to help organize desktop with AI-powered features
Desktop Visualization: View your current desktop layout in a web interface.
AI-Powered Organization: Analyzes your desktop files and automatically suggests appropriate categories/tags using Google Gemini AI. Customizable Layout: Organize your desktop icons efficiently.
Desktop Info Extractor: A C# utility to extract icon positions and images from your actual Windows desktop.

  Intro Page   Code   Demo website