Face Detection Algorithm Based on Lightweight Models
A face detection algorithm based on the single-stage Retinaface network is proposed to address the accuracy issue in face detection algorithms that rely on large-scale parameters.The GhostNet serves as the feature extraction network of the Retinaface network,and the linear computation conserves computational resources and enhances computational efficiency.The information exchanges between feature maps are fortified and the model's feature fusion capabilities are elevated with weighted bidirectional feature pyramid network.The original and optimized models are tested on the Widerface dataset,and experimental results show that the accuracy of optimized model improved by up to 7.35％over the original model.
Single-stage face detection algorithmGhostNetweighted bidirectional feature pyramid network
万方数据
维普
