Jinbo Xing

I am currently a third-year Ph.D student in the CSE Department, CUHK, supervised by Prof. Tien-Tsin WONG.

Previously, I received my B.Sc. (1st class honor & ELITE Stream) and M.Sc. degrees in Computer Science at CUHK in 2020 and 2021 respectively.

I'm focusing on controllable text-to-video & image-to-video generation.

我正在寻找2025年秋季入职的全职岗位(AIGC视频生成), 欢迎邮件联系

I will graduate at 2025 Fall. Drop me an Email if your are recruting!

Email  /  Google Scholar  /  Github  /  ResearchGate  /  LinkedIn  /  CV

arXiv  /  Project  /  Code  /  Demo

An open-domain image animation method based on video diffusion priors.

arXiv  /  Project  /  Code  /  Demo

Given a stylized image as reference, the model can generate images and videos with consistent styles based on text prompt.

arXiv  /  Project  /  Code

An open-sourced foundational text-to-video and image-to-video diffusion model for high-quality video generation.

arXiv  /  Project  /  Code

Given text description and video structure (depth), our approach can generate temporally coherent and high-fidelity videos. Its applications include dynamic 3d-scene-to-video creation, real-life scene to video, and video rerendering.

arXiv  /  Project  /  Code

Given a real-world image, our proposed method can generate its semantic-aligned variants. The approach is compatible with customized image generation, inpainting, controllable image synthesis, style transfer, etc.

arXiv  /  Project  /  Code

We utilize the abundance of existing video clips and synthesize a coherent storytelling video by customizing their appearances with given visual concepts.

arXiv  /  Project  /  Code

We present an effective video colorization method, an adaptation of a pre-trained text-to-image diffusion model for video colorization.

arXiv  /  Project  /  Code

We present a novel image colorization method, which leaverages the rich diffusion model prior. It takes text-based image colorization to another semantic level.

arXiv  /  Paper  /  Project  /  Code  /  Online demo

We minimize the over-smoothed facial expression using a learned discrete motion prior, which means more dramatic and expressive facial motions with more accurate lip movements can be achieved.

arXiv  /  Paper  /  Project  /  Code  /  Video  /  Data

We present a controllable scene stylization method utilizing radiance fields to stylize a 3D scene, with a single stylized 2D view taken as reference.

Paper  /  Project  /  Code  /  Interactive inspection

We present a scale-arbitrary invertible image downscaling network (AIDN) to natively downscale HR images with arbitrary scale factors. Meanwhile, the HR images could be restored with AIDN whenever necessary.

Paper

We present a flow-aware multi-frame interpolation method to address the complicated non-linear motions in the real world.

arXiv  /  Paper  /  Code  /  Paper Reading  /  Poster

We propose a system capable of accessing unstructured knowledge for task-oriented dialog.

Last updated: April 2024
Web page design credit to Jon Barron