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

Abstract

This paper analyses the design choices of face detection architecture that improve efficiency between computation cost and accuracy. Specifically, we re-examine the effectiveness of the standard convolutional block as a lightweight backbone architecture on 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 layer 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, face. Based on our observation, we propose to employ ResNet with a highly reduced channel, which surprisingly allows high efficiency compared to other mobilefriendly networks (e.g., MobileNet-V1,-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 maximizing 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 in on CPU. Code will be available at https://github.com/clovaai/EResFD.

HP-FLDet: A High Performance Joint Face and Landmark Detector

“…From the previous work, there are some observations: (1) Anchor-based face detectors [23,7,17,24,25] have demonstrated the most promising accuracy, particularly on the most challenging dataset Wider Face [26]. (2) Almost all the state-of-the-art(SOTA) anchor-based face detectors adopt multi-scale detection heads and deploy many facial specific modules for dense anchor design [9] and anchor mining [8,23,24], which may lead to anchor misalignment and poor design generality easily.…”