Real-Time Face Detection Algorithm Based on Improved YOLOv8

To resolve the shortcomings of low face detection Hit Rate high computational complexity on face Instance Detection tasks, an improved YOLOv8-based face detection algorithm is proposed. The CBAM attentional component was first fused into the Neck network structure, enhancing The framework's face Identification Correctness while reducing impact of background redundancy on face detection. Secondly, to mitigate the resolution degradation caused by multiple upsampling and downsampling operations, the network of aggregated paths in the Neck is supplanted by a system of weighted bijective pyramid chains. Thirdly, the SPPF module in the Neck is improved, increases face detection accuracy. Finally, to address the model's limitations in recognizing irregularly shaped faces and faces with varying poses, dynamic deformable convolution Embedded within, improving The model's capability for feature extraction from faces with different poses. The model is validated Tested on the widely-used WIDER FACE dataset. On the medium and hard subsets, both detection accuracy and speed are improved. This approach enhances the model's face detection accuracy while main taining good generalization capability.