👁️Vision is the interface where photons meet consciousness - for carbon and silicon alike.

Hi 👋! I’m a CS Ph.D. candidate at UCD in the Vision and Analytics Lab (THEIA Lab), working with Prof. Soumyabrata Dev.

I previously served as a research assistant at SIAT, CAS. collaborating with Prof. Zhengkun Yi, and earned my M.S. degree in Computer Science from USM under the supervision of Prof. Putra Sumari.

My current research focuses on computer vision, machine learning, and AI.

🔥 News

2026.03: 📝 MoCrop, developed under limited compute (single RTX 3090), has been accepted at IEEE IJCNN. paper [code]
2026.02: 📝 MoCLIP-Lite, developed under limited compute (single RTX 3090), has been accepted at IEEE CISP and is now available on IEEE Xplore: paper [code]
2026.01: 📝 TinyDrop has been accepted at IEEE ICASSP!
2025.12: 👨‍💻 Joined an AI startup as an AI Algorithm Developer (Video Agents, Intern)
2025.10: 🤝 Contributed MV-Only (high-speed mode) to mv-extractor. [upstream] [my fork] [PyPI]
2025.09: 📝 Preprinted MoCLIP-Lite on arXiv: paper [code]
2025.09: 📝 Preprinted MVP on arXiv: paper [code]
2025.09: 📝 Preprinted MoCrop on arXiv: paper [code]
2025.07: 👨‍🎓 Obtained a new certification: Deep Learning with PyTorch : Image Segmentation from Coursera!
2025.02: 👨‍🎓 Obtained a new certification: LLM Agents from UC Berkeley!
2024.11: 👨‍🎓 Obtained a new certification: Neural Network and Deep Learning from DeepLearning.AI!
2024.09: 🏆 Ranked top2.5% at National Information Center: Video Recognition for City Management Competition!
2024.08: 🏆 Ranked top3.5% at NVIDIA and Alibaba: Multimodal Large Model Data Synthesis Challenge!
2024.07: 👨‍🎓 Obtained a new certification: Visual Perception from Columbia University!

📝 Latest Publications

2026

MoCrop: Training Free Motion Guided Cropping for Efficient Video Action Recognition
B. Huang, W. Yao, S. Chen, G. Wang, Q. Wang, S. Dev
IEEE IJCNN [2026] [Paper] [Code]
TinyDrop: Tiny Model Guided Token Dropping for Vision Transformers
G. Wang, Q. Wang, B. Huang, S. Chen, D. John
IEEE ICASSP [2026] [Paper]

2025

MoCLIP-Lite: Efficient Video Recognition by Fusing CLIP with Motion Vectors
B. Huang, N. Wang, A. Parakash, S. Dev
IEEE CISP [2025] [Paper] [Code]
MVP: Motion Vector Propagation for Zero-Shot Video Object Detection
B. Huang, N. Wang, W. Yao, S. Dev
arXiv preprint [2025] [Paper] [Code]
Optimal Brain Connection: Towards Efficient Structural Pruning
S. Chen, W. Ma, B. Huang, Q. Wang, G. Wang, W. Sun, L. Huang, D. John
arXiv preprint [2025] [Paper]
DCentNet: Decentralized multistage biomedical signal classification using early exits
X. Li, B. Huang, B. Cardiff, D. John
Biomedical Signal Processing and Control [2025] [Paper] [Code]

2024

Dynamic liquid volume estimation using optical tactile sensors and spiking neural network
B. Huang, S. Fang, M. Yin et al.
Springer - Intelligent Service Robotics [2024] [Paper]
More papers on Google Scholar.

🧩 Open-Source Contributions

mv-extractor — Contributor (2025)
A fast H.264/MPEG-4 motion-vector extractor with optional RGB decode skipping.
• Key: Implemented Motion-Vectors-Only / frame-decode skipping mode; docs & tests; acknowledged upstream.
• Links: Upstream · My fork · PyPI

👨‍💻 Service

Reviewer - AAAI, ACM MM, IEEE ICRA, IEEE ICASSP, IEEE IJCNN, IEEE TIP, IEEE TMM, IEEE T-ASE, IEEE TIM
Technical Program Committee (TPC): IEEE Smart World Congress 2026
Session Assistant - IEEE RCAR 2023

👨‍💻 Research Focuses

 • Video Understanding         (Outputs)       2025 - present
 • Model Compression           (Outputs)       2023 - 2025
 • Embodied Intelligence       (Outputs)       2021 - 2023

🏃‍♂️ Personal life