Hi, it's Sayem, welcome to my Homepage!

Passionate about Vision-Language Models and AI research!

I am a researcher in the Vision and Learning Lab at the Ulsan National Institute of Science and Technology (UNIST), where I also completed my Master's degree in Computer Science and Engineering. I work under the supervision of Prof. Seungryul Baek and Prof. Binod Bhattarai. My research focuses on Vision-Language Models (VLMs) for Hand-Object Mesh Reconstruction and Hand-Object Interaction Understanding.

Previously, I completed my Bachelor's in Industrial Engineering at UNIST. During my undergraduate years, I worked on optimizing machine learning models for human detection using thermal images, improving image segmentation with RGB and thermal fusion, and parallelizing bandit algorithms for computational efficiency in recommendation systems.

My broader research interests span Vision-Language Models (VLMs), 3D Hand-Object Interaction, Multimodal Learning and Scene Understanding.

I am always open to collaborations and discussions. If you are interested in my research or have any inquiries, feel free to reach out to me at khalequzzamansayem@unist.ac.kr.

News

🎉 CVPR 2026: Our paper "HandVQA: Diagnosing and Improving Fine-Grained Spatial Reasoning about Hands in Vision-Language Models" has been accepted to CVPR 2026!
🎉 September 2025: Started my new role as a Researcher at UNIST's Vision and Learning Lab!
🎉 August 2025: I have successfully defended my Master's thesis and graduated from UNIST!
🎉 AAAI 2025: Our paper "QORT-Former: Query-Optimized Real-Time Transformer for Understanding Two Hands Manipulating Objects" has been accepted! Check out our poster!

Publications

HandVQA: Diagnosing and Improving Fine-Grained Spatial Reasoning about Hands in Vision-Language Models. Accepted at CVPR 2026.
Authors: MD Khalequzzaman Chowdhury Sayem, Mubarrat Tajoar Chowdhury, Yihalem Yimolal Tiruneh, Muneeb A. Khan, Muhammad Salman Ali, Binod Bhattarai, Seungryul Baek.
In this paper, we introduce HandVQA, a large-scale benchmark grounded in 3D hand geometry to systematically evaluate fine-grained spatial reasoning in Vision-Language Models (VLMs). The dataset contains 1.6M+ geometry-derived VQA pairs spanning joint angles, distances, and relative spatial relations (X/Y/Z). We demonstrate that explicit 3D supervision significantly improves spatial reasoning reliability and generalizes to gesture recognition and hand–object interaction tasks.
Project Page | Paper (coming soon) | Code (coming soon)
QORT-Former: Query-Optimized Real-Time Transformer for Understanding Two Hands Manipulating Objects. Accepted at AAAI 2025.
Authors: Elkhan Ismayilzada*, MD Khalequzzaman Chowdhury Sayem* (Co-First Author), Yihalem Yimolal Tiruneh, Mubarrat Tajoar Chowdhury, Muhammadjon Boboev, Seungryul Baek.
In this paper, we present a query-optimized real-time Transformer (QORT-Former), the first Transformer-based real-time framework for 3D pose estimation of two hands and an object. Our approach optimizes queries to balance efficiency and accuracy, leveraging hand-object contact information and a three-step feature update mechanism. Our method achieves real-time pose estimation at 53.5 FPS on an RTX 3090TI GPU while outperforming state-of-the-art models on H2O and FPHA datasets.
Project Page | Paper | Code