Wake Vision
收藏资源简介:
Wake Vision是一个专为TinyML人体检测任务设计的大型多样化数据集,由哈佛大学等机构创建。该数据集包含超过600万张图像,比之前的标准数据集大100倍,并经过了严格的质量筛选。数据集的内容涵盖了多种场景和条件,如不同的光照、距离和人群特征,以确保模型在实际环境中的鲁棒性和公平性。创建过程中,数据集从Open Images v7数据集中提取,并根据CC-BY 4.0许可公开发布,支持商业用途。Wake Vision旨在解决TinyML领域中数据集大小和质量限制的问题,推动该领域的研究和应用发展。
Wake Vision is a large-scale diverse dataset specifically designed for TinyML human detection tasks, created by institutions including Harvard University and others. It contains over 6 million images, which is 100 times larger than previous standard datasets, and has undergone rigorous quality filtering. The dataset covers a wide variety of scenarios and conditions, such as varying lighting, shooting distances, and diverse crowd characteristics, to ensure the robustness and fairness of models in real-world deployment scenarios. The dataset was extracted from the Open Images v7 dataset during its creation, and is publicly released under the CC-BY 4.0 license, supporting commercial usage. Wake Vision aims to address the constraints of dataset size and quality in the TinyML domain, and promote the research and application development of this field.
- 1Fast Data Aware Neural Architecture Search via Supernet Accelerated EvaluationTechnical University of Denmark, Microsoft · 2025年



