Face Detection Using Improved Faster RCNN

TL;DR

This paper introduces FDNet1.0, an improved Faster RCNN-based method for face detection that employs multi-scale techniques, ensemble methods, and inference tricks, achieving top results on the WIDER FACE dataset.

Contribution

The paper presents a detailed design of FDNet1.0, enhancing Faster RCNN for face detection with novel techniques and ensemble strategies, leading to state-of-the-art performance.

Findings

Achieved top rankings on WIDER FACE validation dataset

Utilized multi-scale training and testing for improved accuracy

Implemented ensemble methods to boost detection performance

Abstract

Faster RCNN has achieved great success for generic object detection including PASCAL object detection and MS COCO object detection. In this report, we propose a detailed designed Faster RCNN method named FDNet1.0 for face detection. Several techniques were employed including multi-scale training, multi-scale testing, light-designed RCNN, some tricks for inference and a vote-based ensemble method. Our method achieves two 1th places and one 2nd place in three tasks over WIDER FACE validation dataset (easy set, medium set, hard set).