Yuhan Zhu

@YuhanZhu_

CS Ph.D. student at Nanjing University Research Intern at Shanghai AI Laboratory Focus on video-language foundation models

Singapore

zyuhan1999.github.io

Joined October 2022

Following

Followers

Replies Reposts Media

Yuhan Zhu
@YuhanZhu_
Jul 21
🚀 Introducing TimeLens2: Generalist Video Temporal Grounding with Multimodal LLMs. ✨TimeLens2-8B achieves SoTA results across 7 benchmarks, outperforming Qwen3.5-397B-A17B by 5.8 points. Paper: arxiv.org/abs/2607.17423 Model & Data: huggingface.co/collections/MC…
Yuhan Zhu
@YuhanZhu_
Jul 17
Introducing VideoChat3 🐦—a fully open, efficient 4B Video MLLM for general, long-form, and streaming video understanding. We’re releasing the weights, code, data, and training recipes! 🚀 📄 arxiv.org/abs/2607.14935 💻 github.com/MCG-NJU/VideoC… 🌐 mcg-nju.github.io/VideoChat3
Yuhan Zhu
@YuhanZhu_
Jun 4
Excited to share that our paper "FreeRet: MLLMs as Training-Free Retrievers" has been accepted to ICML 2026! 🎉 We show that off-the-shelf MLLMs can serve as powerful multimodal retrievers and rerankers — no extra training needed. 🚀 Paper: arxiv.org/pdf/2509.24621
Yuhan Zhu
@YuhanZhu_
Jun 4
🚀 Excited to share Video-o3, accepted to ICML'26! 🎉 A native clue-seeking framework for long-video multi-hop reasoning 🎥🧠 Instead of watching everything uniformly, Video-o3 actively hunts for key evidence, reasons step by step. Paper 👉 arxiv.org/abs/2601.23224
1
Yuhan Zhu
@YuhanZhu_
Oct 5, 2024
Excited to share that our work on transferring vision-language models has been accepted at NeurIPS 2024! 🎉 📖 Read our preprint: [arXiv](arxiv.org/pdf/2407.04603) 🤖 Check out the code: [GitHub](github.com/MCG-NJU/AWT) Stay tuned for updates! 🚀