Hongwei Xue

I was a researcher at ByteDance Research. Before that, I obtained my Ph.D. degree from University of Science and Technology of China (USTC), My advisors are Jiebo Luo and Houqiang Li. I received the B.S. degree from School of the Gifted Young, USTC.

I spent joyful and fulfilling times at NUS, Tencent WeChat, Shanghai AI Lab, Microsoft Research Asia (MSRA).

Email  /  Google Scholar  /  Github

I'm interested in Multi-Modal Learning, Computer Vision, and Machine Learning. Much of my research is about Vision-and-Language Pre-training. Representative papers are highlighted. (* indicates equal contribution)

We propose a novel parameter space alignment paradigm for MLLMs to address the inefficiency of input space alignment paradigm in visual perception, introducing VLoRA that converts visual features to LoRA weights, achieving comparable performance on various benchmarks while significantly reducing computational costs for training and inference.

We propose a Multi-Modal Generative Embedding Model (MM-GEM), whereby the generative and embedding objectives are encapsulated in one Large Language Model. we explore the minimalism of multi-modal paradigm in this work.

we propose an efficient MIM paradigm named MaskAlign. MaskAlign simply learns the consistency of visible patch features extracted by the student model and intact image features extracted by the teacher model.

We adapt image-text pre-trained models to video-text pre-training (i.e., post-pretraining). In this work, we propose an Omnisource Cross-modal Learning method equipped with a Video Proxy mechanism on the basis of CLIP, namely CLIP-ViP.

We introduce a Long-Form VIdeo-LAnguage pre-training model (LF-VILA) and train it on a large-scale long-form video and paragraph dataset constructed from HD-VILA-100M.

We collect a large dataset which is the first high-resolution dataset including 371.5k hours of 720p videos and the most diversified dataset covering 15 popular YouTube categories.

We propose a fully Transformer model for Vision-and-Language pre-training and explore to study the inter-modal interaction.

We propose a model named FGLA to generate high-quality and realistic videos by learning Fine-Grained motion embedding for Landscape Animation.

In this work, we propose a unified image-and-text generative framework based on a single multimodal model to jointly study the bi-directional tasks.

We propose to leverage the semantic tags to bridge the gap between the modalities of vision and language rather than directly concatenating or attending to the visual and linguistic features.

We propose a deep Siamese network model to to classify different types of image edits between an original image and an edited image.

• Reviewer of ACM MM, ICMR, ICME, TMM, CVPR, ICCV.
• National Scholarship at USTC.
• Scholarship for Excellent Students and Freshmen Scholarship at USTC.

Based on Jon Barron's website.