基于空间自适应Transformer的人脸素描图像生成

万伟国,钟凯晨,张迎梅,姚丽,包书鸿,龚大伟

Face photo-to-sketch synthesis based on spatially adaptive Transformer

Weiguo WAN,Kaichen ZHONG,Yingmei ZHANG,Li YAO,Shuhong BAO,Dawei GONG

表 1 不同方法在CUHK、AR、XM2VTS和CUFSF数据集上的定量指标对比

Tab.1 Comparison of quantitative metrics of different methods on CUHK, AR, XM2VTS, and CUFSF datasets

模型 CUHK数据集 AR数据集 SSIM PSNR/dB MSE LPIPS FID SSIM PSNR/dB MSE LPIPS FID pix2pix 0.6587 30.70 56.30 0.1731 89.08 0.5946 29.68 71.83 0.2150 99.92 U-GAT-IT 0.5551 29.75 69.28 0.3206 131.85 0.4734 29.28 77.79 0.4122 170.04 DCLGAN 0.6479 30.48 58.56 0.2149 72.46 0.6128 29.78 70.46 0.1893 97.96 文献[15]模型 0.6506 30.12 64.26 0.1893 71.27 0.6047 29.80 69.31 0.2180 82.45 CAP-VSTNet 0.6389 30.43 59.34 0.2228 82.41 0.5778 29.65 71.06 0.2710 117.35 Mamba-ST 0.5365 29.69 72.03 0.3460 75.40 0.6041 29.06 81.17 0.2379 84.68 SAT-StyTR 0.6676 30.89 53.93 0.1745 57.51 0.6213 29.89 67.88 0.2043 81.66 模型 XM2VTS数据集 CUFSF数据集 SSIM PSNR/dB MSE LPIPS FID SSIM PSNR/dB MSE LPIPS FID pix2pix 0.4527 29.22 78.11 0.2514 62.13 0.3615 28.36 94.94 0.2770 54.89 U-GAT-IT 0.4359 29.30 76.70 0.4471 137.14 0.3981 28.27 96.95 0.2731 42.63 DCLGAN 0.4983 29.22 78.28 0.2847 91.56 0.3750 28.35 95.12 0.2584 44.22 文献[15]模型 0.4978 29.47 73.95 0.2732 63.65 0.3823 28.25 97.26 0.2725 32.56 CAP-VSTNet 0.4968 29.23 78.15 0.2710 79.81 0.3655 28.10 100.65 0.2909 61.21 Mamba-ST 0.4536 28.37 94.81 0.3449 74.16 0.4015 28.22 98.01 0.2818 49.03 SAT-StyTR 0.5116 29.52 73.19 0.2436 52.48 0.4236 28.51 91.74 0.2356 29.30