Xiaobin Hu (胡晓彬)

I am currently a Senior Research Fellow at the LV-Lab of National University of Singapore, under the supervision of Prof. Shuicheng Yan. Before that, I obtained my Ph.D. degree at the School of Computer Science and Engineering, Technische Universität München, Germany, under the joint supervision of Prof. Bjoern Menze and Prof. Kuangyu Shi.

  Email  /    Google Scholar  /    Github  / 

News

• 04/2026 – 🎉 The Latent Space: Foundation, Evolution, Mechanism, Ability, and Outlook is released. Recognized as the #2 Paper of the Day on Hugging Face, we are happy to release this comprehensive survey and open-source repository defining the evolving landscape of latent space research. This work charts a critical paradigm shift in language-based models: moving away from the structural limitations of explicit token-level generation toward highly efficient, continuous latent space computation. Read the full paper on arXiv.

• 02/2026:  🎉🎉🎉 VisMem, MACT, Soul, Replay, DiP, Med-CMR, FFP-300K, KeyTailor, Internal Guidance are accepted by CVPR 2026.

• 01/2026  🎉🎉🎉 Human-MME, ViF, IVEBench, M3CoTBench, SAD, MAS2 are accepted by ICLR 2026.

• 12/2025 – I was invited as Area chair for ECCV 2026.

• 11/2025 – The Landscape of Medical Agents: A Survey is released. Collaborating with renowned experts from TUM, NUS, Harvard, and Fudan, we are proud to release "The Landscape of Medical Agents: A Survey," complete with open-source resources to define this emerging field. This work charts the critical evolution of medical AI from static, passive tools to active, context-aware agents capable of reasoning, planning, and seamlessly integrating into clinical workflows.

• 10/2025 – PointSeg honored with the ICCV workshop Best Demonstration Award

• 07/2025 – IPVG accepted by ACM MM 2025 and attain the runner-up in the 2025 ACM Multimedia Challenge (Identity-preserving video generation challenge)

• 07/2025 – DICE-Talk and StrandDesigner accepted by ACM MM 2025

• 06/2025 – OracleFusion and UniCombine accepted by ICCV 2025

• 02/2025 – Sonic , VTON-HandFit , FTEdit , CustAny , GroundingFace , SVFR , DVHGNN , Mobilemamba accepted by CVPR 2025, ranked 92th globally.

• 07/2024 – 3Diffusion accepted by MM 2024

• 07/2024 – RLR and DiffuMatting accepted by ECCV 2024

• 07/2024 – 1 paper accepted by IEEE Transactions on Circuits and Systems for Video Technology (TCSVT).

• 05/2024 – 1 paper accepted by Pattern Recognition (PR).

We propose a novel paradigm, dubbed as Sonic, to shift focus on the exploration of global audio perception.

We propose OracleFusion, a novel two-stage semantic typography framework for generating semantically rich vectorized fonts that accurately convey the essence of the original glyphs, providing a structured visual reference for oracle character interpretation.

Although diffusion-based image virtual try-on has made considerable progress, emerging approaches still struggle to effectively address the issue of hand occlusion (i.e., clothing re- gions occluded by the hand part), leading to a notable degradation of the try-on performance. To tackle this issue widely existing in real-world scenarios, we propose VTON-HandFit, leveraging the power of hand priors to reconstruct the appearance and structure for hand occlusion cases.

FitDiT is designed for high-fidelity virtual try-on using Diffusion Transformers (DiT).

The prevailing diffusion inversion performs deficiently in flow-based models, and the invariance control cannot reconcile diverse rigid and non-rigid editing tasks. To address these, we systematically analyze the inversion and invariance control based on the flow transformer.

SVFR is a unified framework for face video restoration that supports tasks such as BFR, Colorization, Inpainting, and their combinations within one cohesive system.

To build the highly accurate Dichotomous Image Segmentation dataset (DIS5K), we first manually collected over 12,000 images from Flickr1 based on our pre-designed keywords. Then, we obtained 5,470 images of 22 groups and 225 categories from the 12,000 images according to the structural complexities of the objects. Each image is then manually labeled with pixel-wise accuracy using GIMP. The labeled targets in DIS5K mainly focus on the “objects of the images defined by the pre-designed keywords (categories)” regardless of their characteristics e.g., salient, common, camouflaged, meticulous, etc. The average per-image labeling time is ∼30 minutes and some images cost up to 10 hours.

Our DiffuMatting shows several potential applications (e.g., matting-data generator, community-friendly art design and controllable generation).

Video deblurring is a challenging task as the blur is often spatially variant. Existing methods mainly engage in building the spatial-temporal correspondence among the frames

A customized multi-level feature extraction method is employed to exploit both structural and identity information of 3D facial images, which are then mapped into the noise estimation process.

To tackle this challenge, we aim to extract the high-resolution texture details to avoid the detail degradation that causes blurred vision in edges and boundaries.

In this study, we present a novel MRI synthesizer, called AutoGAN-Synthesizer, which automatically discovers generative networks for cross-modality MRI synthesis.

Existing face restoration algorithms only employ 2D priors without considering high dimensional information (3D). The 3D morphable facial priors are the main novelty of this work and are completely different from recently related 2D prior works

In this paper, we propose a novel face super resolution method that explicitly incorporates 3D facial priors which grasp the sharp facial structures. Our work is the first to explore 3D morphable knowledge based on the fusion of parametric descriptions of face attributes (e.g., identity, facial expression, texture, illumination, and face pose)

we propose a novel deblurring method, dubbed PyNAS, towards automatically designing hyper-parameters including the scales, patches, and standard cell operators. Our primary contribution is a real-time deblurring algorithm (around 58 fps) for 720p images while achieves state-of-the-art deblurring performance on the GoPro and Video Deblurring datasets.