Xinghao Chen陈星昊 陈星昊

Research

1. ViTeX-Bench: Benchmarking High-Fidelity Video Scene Text Editing

   Xinghao Chen, Xiangbo Gao, Jiongze Yu, Yuheng Wu, Zhengzhong Tu

   NeurIPS 2026 Evaluations & Datasets Track First author

   Replace First Last SOC COC ONLY STOP COLLIER WASHING

   $t=$0

   Loading…

   Video scene text editing replaces the words on signs, boards and labels in a real video while the rest of the scene and its motion stay intact. ViTeX-Bench pairs ViTeX-Dataset, 387 real-world 720p videos, with 13 metrics over text correctness, visual and temporal quality, and edit locality, and releases ViTeX-Edit-14B as an open reference editor.

   There is no single score. Each editor is a point on one primary metric per axis, and the front is the set no other editor beats on all three: across eight baselines from four editing families, accurate text, temporal stability and scene preservation remain hard to get together.

   • Project page
   • Code
   • Dataset
   • Model
   • Leaderboard

2. PISCO: Precise Video Instance Insertion with Sparse Control

   Xiangbo Gao, Renjie Li, Xinghao Chen, Yuheng Wu, Suofei Feng, Jie Yang, Qing Yin, Zhengzhong Tu

   NeurIPS 2026 Main Track

   Inserting an object into an existing video from a few keyframes at any timestamps. PISCO carries its appearance, motion, shadows and reflections through the clip while the original scene and its dynamics stay intact.

   • Paper
   • Project page
   • Code
   • Model

   Insert Desk lamp Rowboat Bear

   $t=$0

3. PVIR-Bench: A Physics-Aware Benchmark for Video Instance Removal

   Zirui Li, Xinghao Chen, Lingyu Jiang, Xiangbo Gao, Dengzhe Hou, Kazunori Yamada, Fangzhou Lin, Zhengzhong Tu

   CVPR 2026 Workshop

   Removing an object from a video along with the shadows, reflections and other physical effects it leaves behind. It holds 95 real videos with instance masks and removal prompts, split into Simple and Hard and rated by people on instruction following, rendering quality and edit exclusivity.

   • Paper

   

   Remove Duck Go-kart Dancer

   $t=$0

Projects

1. Community photo

   Segmentation

   CoastalSeg

   Image segmentation for coastal erosion monitoring

   2025 · University of Washington Applied Physics Laboratory · capstone project

   A system for multi-class segmentation of shoreline photos uploaded by community members, with outlier detection and multi-image perspective correction. The segmentation model, DeepLabV3+ with an EfficientNet-B6 encoder, reaches 0.93 IoU. Built for coastal research at the UW Applied Physics Laboratory and used with MyCoast Washington.

   With Zheheng Li, Dylan Scott, Aaryan Shah, Bauka Zhandulla and Sarah Li.

   Semantic segmentation · Outlier detection · Perspective alignment

   • Code
   • Demo

2. The UW campus in VR

   A walkable campus tour for Meta Quest

   2025 · Developing Immersive Experiences for AR/VR, University of Washington

   I photographed landmark buildings and sculptures around the University of Washington with a phone, reconstructed them as 3D Gaussian splats, and assembled them into a campus tour you can walk through in a Meta Quest headset.

   3D Gaussian Splatting · Meta Quest · Phone capture

News

1. Spring 2027

   Joining Texas A&M University as a Ph.D. student in Computer Science, in the TACO Group advised by Dr. Zhengzhong Tu.

2. Sep 24, 2026

   ViTeX-Bench, the project I led as first author, was accepted to the NeurIPS 2026 Evaluations & Datasets Track.

3. Sep 24, 2026

   PISCO, led by Xiangbo Gao with me as a co-author, was accepted to the NeurIPS 2026 main track. Congratulations, Xiangbo!

4. Dec 2025

   Graduated from the University of Washington with an M.S. in Electrical & Computer Engineering.

Visitors