|
Xiaojuan Wang I'm a fifth-year Ph.D student in GRAIL, University of Washington, advised by Steve Seitz, Brian Curless and Ira Kemelmacher. I also work closely with Aleksander Holynski. Previously, I received my Master's degree from ETH Zurich, working in the CVG lab. I have also interned at Meta, Adobe Research, and Google Research during summer times.
I am broadly interested in visual generative models, especially video models, and how to use these models to help video content creation in real world applications. When I am not doing research, I read and play tennis. |
|
Selected Publications |
|
Generative Inbetweening: Adapting Image-to-Video Models for Keyframe Interpolation
Xiaojuan Wang, Boyang Zhou, Brian Curless, Ira Kemelmacher, Aleksander Holynski, Steve Seitz arXiv, 2024 project page / arXiv Given a pair of key frames as input, our method generates a continuous intermediate video with coherent motion by adapting a pretrained image-to-video diffusion model. |
|
Generative Powers of Ten
Xiaojuan Wang, Janne Kontkanen, Brian Curless, Steve Seitz, Ira Kemelmacher, Ben Mildenhall, Pratul P. Srinivasan, Dor Verbin, Aleksander Holynski CVPR, 2024 (Highlight) project page / arXiv Given a series of prompts describing a scene at drastically varying scales, our method creates a seamless zooming video. |
|
Jump Cut Smoothing for Talking Heads
Xiaojuan Wang, Taesung Park, Yang Zhou, Eli Shechtman, Richard Zhang arXiv Preprint, 2023 project page / arXiv Given a talking head video, we remove the the filler words, repetitive words and so on, and create a seamless transition for the jump cut. |
Some Challenging Projects I Explored |
| • Photorealistic NBA2K; Xiaojuan Wang, Brian Curless, Ira Kemelmacher, Steve Seitz |
| TL;DR: Given a synthetic NBA2K game video (left), we turn it into a corresponding real NBA video (Right), i.e., the court and all players have realistic appearances. Challenge: we do not have such video pairs to directly learn the translation. |
Earlier Works |
|
• Learning 3D semantic reconstruction on octrees (GCPR 2019); Xiaojuan Wang, Martin Oswald, Ian Cherabier, Marc Pollefeys
• Supervised quantization for similarity search (CVPR 2016); Xiaojuan Wang, Ting Zhang, Guo-Jun Qi, Jinhui Tang, Jingdong Wang • Multi-scale learning for low-resolution person re-identification (ICCV 2015); Xiang Li, Wei-Shi Zheng, Xiaojuan Wang, Tao Xiang, Shaogang Gong • Cross-scenario transfer person reidentification (TCSVT 2015); Xiaojuan Wang, Wei-Shi Zheng, Xiang Li, Jianguo Zhang |
Miscellanea |
|
My first name in Chinese is 小鹃，which means "a little cuckoo bird". You can also call me "jeanne".
I travel a lot, and enjoy spending time in the musuem. |
|
This website borrows the template from Jon Barron |