NEWS

2025.10.2

AWL’s Research Paper “PITR-Select” Accepted for Presentation at VCIP2025 – Breakthrough in Efficient Edge-Based Video Understanding

• NEWS
• 研究成果・発表

AWL Inc.’s research paper, “PITR-Select: Partial Image Token Reduction via Temporal Selection for Efficient Video Understanding on Edge”, has been accepted to VCIP2025 (Visual Communications and Image Processing), an international conference technically sponsored by IEEE CAS.
The paper introduces PITR-Select, a novel algorithm designed to optimize the performance of Vision Language Models (VLMs) in edge computing environments. This technology forms the foundation of AWL’s upcoming General AWL engine, which aims to enhance real-time video analysis while significantly reducing power consumption.
AWL will continue to accelerate real-world implementation of this research to help solve pressing societal challenges.

*This result is based on results obtained from a project, JPNP23019, subsidized by the New Energy and Industrial Technology Development Organization (NEDO).