Attack image generation method, training method of generative network and related device

The invention discloses an attack image generation method, a training method of a generative network and a related device. The attack image generation method comprises the following steps: acquiring an attacker image; based on the attacker image, an original attack image matched with the attacked person image in style is generated, and the original attack image comprises image content of a corresponding attack area in the attacker image; the original attack image is mapped to a printer color space, a target attack image is obtained, and the target attack image is used for forming attack materials for attacking the biometric authentication system after being printed. According to the scheme, the problem that the digital color space is not matched with the printer color space in the adversarial attack test process can be solved, so that the effectiveness of physical attacks is improved. 本申请公开了攻击图像生成方法及生成网络的训练方法、相关装置，攻击图像生成方法包括：获取攻击者图像；基于攻击者图像，生成风格与被攻击者图像匹配的原始攻击图像，原始攻击图像包含攻击者图像中对应攻击区域的图像内容；将原始攻击图像映射至打印机颜色空间，得到目标