Twitxy Saved

Xiaohan Wang

Verified account This account is verified. Learn more

@XiaohanWang96

Postdoc @Stanford. Video Understanding, Multimodal Foundation Models
Palo Alto, CA ai.stanford.edu/~xhanwang/ Joined November 2016
482 Following 322 Followers 79 Posts

Latest posts 5

Xiaohan Wang Verified account @XiaohanWang96
HUGE day for video understanding. 🔥

Agentic video is a product now, and the core idea is the one we introduced in VideoAgent (Mar 2024): let the model observe, think, and decide which frames to look at next.

This is why I love this field. Nothing beats watching your idea become real at the frontier of video intelligence. @Zhang_Yu_hui @orr_zohar @yeung_levy

Congrats to the Gemini team 🙌

arxiv.org/abs/2403.10517
arxiv.org VideoAgent: Long-form Video Understanding with Large Language... Long-form video understanding represents a significant challenge within computer vision, demanding a model capable of reasoning over long multi-modal sequences. Motivated by the human cognitive...
Google Verified account @Google Sep 1
We’re introducing a new capability to our latest Gemini models: agentic video understanding.

This allows developers to process long-form video content with more accuracy, while using up to 88% less tokens.

See how it works 🧵
View quoted post on X
2 replies 4 reposts 16 likes 845 views View on X
Xiaohan Wang Verified account @XiaohanWang96
🚀 Excited to release SciVideoBench — a new benchmark that pushes Video-LMMs to think like scientists!

Designed to probe video reasoning and the synergy between accurate perception, expert knowledge, and logical inference.

1,000 research-level Qs across Physics, Chemistry, Biology & Medicine, testing conceptual, hypothetical, and quantitative understanding.

Let’s see which models can truly reason in science 🔬
Shoubin Yu Verified account @shoubin621 Oct 10, 2025
🚨 New Paper Alert! Introducing SciVideoBench — a comprehensive benchmark for scientific video reasoning!

🔬SciVideoBench:

1. Spans Physics, Chemistry, Biology & Medicine with authentic experimental videos.

2. Features 1,000 challenging MCQs across three reasoning types: Conceptual, Hypothetical, and Quantitative.

3. Explicitly targets higher-order multimodal cognitive skills in LMMs.

4. Provides valuable insights on Chain-of-Thought prompting, scaling effects, and reasoning limits.
🧵
View quoted post on X
Xiaohan Wang Verified account @XiaohanWang96
🧠 How can we truly test long-context video understanding in video-LMMs?

⏱️ TimeScope benchmarks models from 1 min to 8 hours using “needle-in-a-haystack” probes.
🚀 Gemini 2.5-Pro leads the pack—but even it struggles as context length grows. Long-range memory is still a frontier.
Orr Zohar Verified account @orr_zohar Jul 23, 2025
🧵 Introducing TimeScope, an open-source benchmark rigorously evaluating the true “temporal context window” of video-language models on videos ranging from 1 minute to 8 hours. #AI #MachineLearning
View quoted post on X
1 replies 1 reposts 10 likes 1.3K views View on X
Xiaohan Wang Verified account @XiaohanWang96
🚨 Excited to co-organize our @CVPR workshop on "Multimodal Foundation Models for Biomedicine: Challenges and Opportunities" — where vision, language, and health intersect!
We’re bringing together experts from #CV, #NLP, and #healthcare to explore:
🧠 Technical challenges (e.g. cross-modal alignment, modality-specific context)
🚀 New opportunities (e.g. multi-scale diagnosis, agentic systems)
We invite short, non-archival submissions (max 4 pages) on:
1️⃣Multimodal foundation models
2️⃣Agentic frameworks
3️⃣Benchmarks & failure modes
4️⃣Novel biomedical applications
🗓 Submit by April 30
📍 See you in Nashville at #CVPR2025!
Workshop date: 1:30-5:30 PM, June 11, 2025
Workshop website: sites.google.com/view/mmfm-biom…
0 replies 16 reposts 58 likes 9.4K views View on X
Xiaohan Wang Verified account @XiaohanWang96
🚀 Introducing Temporal Preference Optimization (TPO) – a video-centric post-training framework that enhances temporal grounding in long-form videos for Video-LMMs! 🎥✨

🔍 Key Highlights:
✅ Self-improvement via preference learning – Models learn to differentiate well-grounded from inaccurate responses without manual annotations.

✅ Multi-level temporal grounding – Effectively captures both localized segments and comprehensive video sequences.

✅ Efficient and scalable – Utilizes only 10k video QA pairs for post-training and can scale seamlessly to larger datasets.

✅ Proven performance – TPO enhances two state-of-the-art Video-LMMs on LongVideoBench, MLVU, and Video-MME.

🔥 LLaVA-Video-TPO is now the top-performing 7B model on Video-MME, highlighting TPO's potential in advancing temporal reasoning.

This work was co-led with the talented undergraduate @rui__li, alongside fantastic collaborators @Zhang_Yu_hui and Zeyu Wang, and advised by @yeung_levy!