Explore cutting-edge research on enhancing energy efficiency and robustness in tinyML computer vision applications through a 25-minute talk from the tinyML Research Symposium 2022. Delve into Qianyun LU's presentation on utilizing log-gradient input images to improve CV pipelines for microcontrollers. Learn about conventional CV pipelines, the process of computing log from RAW images, and gain intuition on energy breakdowns. Discover insights from CNN experiments, dataset considerations, and architecture search using UNAS for microcontrollers. Examine the impact on robustness to illumination changes and grasp key takeaways in this comprehensive overview of innovative tinyML techniques.
Improving Energy Efficiency and Robustness of tinyML Computer Vision Using Log-Gradient Input Images
Overview
Syllabus
Intro
Conventional computer vision (CV) pipeline
Compute log from RAW
Intuition
Energy breakdown of pipelines
Overview of CNN experiments
Datasets
Architecture search: UNAS for microcontrollers
Robustness to illumination change
Summary
Taught by
tinyML
