Aggregating Deep Features for Image Retrieval

In this article, we propose a new approach to object detection on mobile devices that improves accuracy while reducing computing power consumption. Our proposed framework leverages a novel offloading strategy that allows the device to transfer the neural network’s weights to a remote server for inference, reducing the computational load on the device’s processing unit. We evaluate our approach using several experiments and demonstrate its effectiveness in achieving accurate object detection while minimizing power consumption.
Our proposed framework consists of three main components: (1) a local neural network that processes the input image and outputs a set of feature maps, (2) an offloading strategy that transfers the weight of the neural network to a remote server for inference, and (3) a data fusion method that combines the output from the local and remote neural networks. By using this framework, we can reduce the computational load on the device’s processing unit while maintaining accuracy, making it ideal for mobile devices with limited computing resources.
We evaluate our approach using several experiments, including object detection in different scenarios and varying levels of computing power consumption. Our results show that our proposed framework outperforms existing approaches in terms of both accuracy and energy efficiency. Specifically, we demonstrate that our offloading strategy can significantly reduce the computational load on the device’s processing unit without sacrificing accuracy, making it an ideal solution for mobile devices.
In conclusion, our proposed framework offers a novel approach to object detection on mobile devices that balances accuracy and computing power consumption. By leveraging a remote server for inference, we can significantly reduce the computational load on the device’s processing unit while maintaining accurate object detection. This makes it an ideal solution for mobile devices with limited computing resources, allowing them to perform complex tasks such as object detection without sacrificing battery life.