Seon-Ho Lee

Title: Seon-Ho Lee and Long-Form Video Understanding.

Introduction

Seon-Ho Lee is an inventor associated with Seattle, Washington, and Amazon Technologies, Inc. His patent work addresses a challenge in video analysis: efficiently processing long-form videos so they can be classified.

Latest Patents

Lee is associated with 1 patent, titled “Merging Tokens for Long-Form Video Understanding.” The invention divides images from a long-form video into tokens or patches, then uses transformer models to merge selected tokens for video classification. The tokens may be partitioned uniformly, according to their locations or motion, or through a learned model that estimates their saliency. After source tokens are matched with target tokens, the resulting merged tokens are processed to classify the video file.

Career Highlights

Lee’s listed patent reflects work at the intersection of transformer models and video understanding. Its approach aims to reduce or organize the visual information processed from long-form video while retaining information useful for classification.

Collaborations

Lee’s listed coworker is David Jiatian Fan. The patent information provided identifies Amazon Technologies, Inc. as Lee’s company.

Conclusion

Seon-Ho Lee’s patent describes a token-merging approach for applying transformer models to long-form video classification, highlighting a practical innovation in video understanding.