EResFD: Rediscovery of the Effectiveness of Standard Convolution for Lightweight Face Detection

TL;DR

This paper demonstrates that heavily channel-pruned standard convolutional blocks in ResNet can outperform depthwise separable convolutions in lightweight face detection, offering better accuracy and speed.

Contribution

The study re-evaluates the effectiveness of standard convolution in lightweight face detection and proposes a ResNet-based backbone with reduced channels for improved efficiency.

Findings

Achieves 80.4% mAP on WIDER FACE Hard subset

Runs inference at 37.7 ms for VGA images on CPU

Outperforms MobileNet variants in accuracy and speed

Abstract

This paper analyzes the design choices of face detection architecture that improve efficiency of computation cost and accuracy. Specifically, we re-examine the effectiveness of the standard convolutional block as a lightweight backbone architecture for face detection. Unlike the current tendency of lightweight architecture design, which heavily utilizes depthwise separable convolution layers, we show that heavily channel-pruned standard convolution layers can achieve better accuracy and inference speed when using a similar parameter size. This observation is supported by the analyses concerning the characteristics of the target data domain, faces. Based on our observation, we propose to employ ResNet with a highly reduced channel, which surprisingly allows high efficiency compared to other mobile-friendly networks (e.g., MobileNetV1, V2, V3). From the extensive experiments, we show that the proposed backbone can replace that of the state-of-the-art face detector with a faster inference speed. Also, we further propose a new feature aggregation method to maximize the detection performance. Our proposed detector EResFD obtained 80.4% mAP on WIDER FACE Hard subset which only takes 37.7 ms for VGA image inference on CPU. Code is available at https://github.com/clovaai/EResFD.