# Accelerating Proposal Generation Network for \\Fast Face Detection on Mobile Devices

## TL;DR

This paper introduces an acceleration framework for proposal generation in face detection, significantly improving inference speed on mobile devices while maintaining high accuracy, by reducing image pyramid levels using facial features.

## Contribution

The paper proposes a novel acceleration method that reduces pyramid levels in cascaded CNNs for face detection, enabling real-time performance on mobile devices.

## Key findings

- Achieved faster inference speed compared to state-of-the-art methods.
- Maintained comparable detection accuracy on benchmarks.
- Demonstrated effectiveness on WIDER-face and FDDB datasets.

## Abstract

Face detection is a widely studied problem over the past few decades. Recently, significant improvements have been achieved via the deep neural network, however, it is still challenging to directly apply these techniques to mobile devices for its limited computational power and memory. In this work, we present a proposal generation acceleration framework for real-time face detection. More specifically, we adopt a popular cascaded convolutional neural network (CNN) as the basis, then apply our acceleration approach on the basic framework to speed up the model inference time. We are motivated by the observation that the computation bottleneck of this framework arises from the proposal generation stage, where each level of the dense image pyramid has to go through the network. In this work, we reduce the number of image pyramid levels by utilizing both global and local facial characteristics (i.e., global face and facial parts). Experimental results on public benchmarks WIDER-face and FDDB demonstrate the satisfactory performance and faster speed compared to the state-of-the-arts. %the comparable accuracy to state-of-the-arts with faster speed.