Introducing Wake Vision: A High-Quality, Large-Scale Dataset for TinyML Computer Vision Applications
Wake Vision is a new large-scale dataset created to advance research and development in TinyML, which focuses on running machine learning models on low-power devices like microcontrollers. The dataset contains approximately 6 million images, nearly 100 times larger than the previous Visual Wake Words (VWW) dataset. Wake Vision offers high-quality labeled images, beneficial for under-parameterized models, and includes fine-grained benchmarks for real-world applications. The dataset is freely available under a permissive license, aiming to help researchers build better person detection models for ultra-low-power devices.
Table of contents
Why TinyML Needs Better DataWhat Makes Wake Vision Different?Why Data Quality Matters for TinyML ModelsReal-World Testing: Wake Vision's Fine-Grained BenchmarksKey Performance Gains With Wake VisionWake Vision Leaderboard: Track and Submit New Top-Performing ModelsMaking Wake Vision Easy to AccessSort: