🚀 Research Homepage | Jongseo Lee

[2026. 07] Seeking PhD Positions in the U.S. (Fall 2027)! 🎓

I completed my M.S. in Computer Science at Kyung Hee University and am now continuing my research as a Post-Master Researcher at the Vision and Learning Lab.

I am actively looking for PhD positions in the United States for Fall 2027, focusing on video understanding, multimodal AI, and explainability. If you think I could be a good fit for your group, I would love to hear from you — feel free to reach out at jong980812@khu.ac.kr!

[2026. 05] New Preprint on arXiv! 🧭

We released a new preprint, “Which Way Did It Move? Diagnosing and Overcoming Directional Motion Blindness in Video-LLMs”.

We identify a fundamental failure mode of Video-LLMs — directional motion blindness — and introduce DeltaDirect, a parameter-efficient motion-change head that raises LLaVA-Video-7B from 27.6% to 85.4% on real-world direction QA.

📄 Check it out on arXiv (2605.22823) and the project page.

[2025. 11. 07] IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) Paper Accepted! 🏆

I’m honored to share that our paper has been accepted to IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), one of the most prestigious journals in computer vision and AI!

Title: “CA²ST: Cross-Attention in Audio, Space, and Time for Holistic Video Recognition”

This work extends our NeurIPS 2023 paper CAST by incorporating audio-visual reasoning into a unified cross-attention framework, enabling models to understand how sound, spatial cues, and temporal dynamics interact in complex video scenes.

We are proud that this work has been recognized by TPAMI — a meaningful milestone in our ongoing exploration of multimodal video understanding.

📄 Check it out on arXiv (2503.23447).

🙏 Huge thanks to my amazing collaborators — Joohyun Chang and Jinwoo Choi† — for their constant guidance and dedication throughout this project.

[2025. 09. 19] NeurIPS 2025 Spotlight Paper Accepted! 🌟

I’m beyond excited to announce that our paper has been accepted to NeurIPS 2025 as a Spotlight (3.5% acceptance rate), one of the most prestigious venues in machine learning and AI!

Title: “Disentangled Concepts Speak Louder Than Words: Explainable Video Action Recognition”

This work introduces a novel framework for structured and interpretable video action recognition, disentangling motion dynamics, objects, and scenes into human-understandable concepts. By doing so, it provides not only strong performance but also clear explanations of model decisions.

The Spotlight recognition makes this achievement even more meaningful, as only a small fraction of submissions receive this honor. It’s an encouraging step forward in my research journey on explainable AI for video understanding.

🙏 I’m truly grateful to my amazing collaborators — Wooil Lee, Gyeong-Moon Park†, Seong Tae Kim†, and Jinwoo Choi† — for their invaluable contributions. Excited to present this work at NeurIPS and share our vision for more transparent and interpretable AI systems!

[2025. 07. 24] ICCV 2025 Highlight Paper Accepted! 🎉

I’m thrilled to share that my paper has been accepted to ICCV 2025 as a highlight paper, one of the top-tier conferences in computer vision!

Title: “ESSENTIAL: Episodic and Semantic Memory Integration for Video Class-Incremental Learning”

What makes this even more meaningful is that it was accepted after three submission attempts. The journey was challenging — filled with revisions, rejections, and countless hours of iteration — but it proved to be an incredible learning experience.

I’m deeply grateful for the growth that came through the process and couldn’t be more excited to finally see it accepted.

🙏 Thank you to everyone who supported me along the way. I look forward to presenting this work at ICCV and continuing to push forward in research!