PhD in Compressed Sensing for Egocentric Video

自从深度神经网络（DNN）出现以来，计算机视觉已经得到了极大的丰富。这些改进主要以卷积神经网络[2]（CNN）或最近的变压器[11]的形式出现。无论模型的形式如何，这些技术都主导着旨在自动理解图像和视频的领域。计算机视觉的关键挑战之一是确保模型能够用方便的硬件（轻便且不显眼）实现，以实时解决问题。在可穿戴技术的情况下，这些物理限制更加复杂，目前的硬件通常笨重，庞大或昂贵。

Computer vision has enjoyed much enrichment since the dawn of deep neural networks (DNNs). These improvements have largely arrived in the form of convo- lutional neural networks [2] (CNNs) or more recently, transformers [11]. Regardless of the form of the model, these techniques dominate fields where the aim is to auto- mate image and video understanding. One of the key challenges in computer vision is ensuring that the model is able to be implemented with convenient hardware (light and unobtrusive) that solves the problem in real-time. These physical constraints are compounded in the case of wearable technology, where current hardware is often unwieldy, large or expensive.