Swin-MFA: A Multi-Modal Fusion Attention Network Based on Swin-Transformer for Low-Light Image Human Segmentation.