USING MACHINE LEARNING TO ESCAPE THE UNCANNY VALLEY

随着数字艺术的兴起，恐怖谷已经从一个深奥的机器人概念变成了一种具有感染力的模因现象，特定的模因，如“不可思议/精明的不可思议先生”，或者更普遍的神秘面孔被用作幽默的反应图像。评论家和玩家现在会将特定的媒体称为“不可思议”，而不是使用更一般的词作为“off-putting”，随着越来越多的例子，展示了在公众意识中巩固自己的神秘感;因此数字艺术家应该意识到即使使用现代渲染技术也能唤起神秘感，因为观众对神秘感的辨别力越来越高。这在视频游戏中最为相关，因为渲染是实时执行的，这意味着必须实现渲染约束。这可能会将角色限制在恐怖谷，因为它可能无法提高图形保真度，因此艺术家可能会接受恐怖或拆除他们的作品（两种不受欢迎的选择）。该项目旨在了解与现代皮肤渲染技术有关的神秘谷，使用人工智能（特别是GAN）将皮肤渲染参数直接映射到用户对神秘感和现实感的评估上。然后，这可以被逆向工程以提供自动化工具，用于生成性地渲染逼真的非神秘皮肤，并预测观众对皮肤真实性的反应，加速QA测试。主要的实验阶段是生成一个具有真实感皮肤的人脸数据库，由参与者使用心理测量学进行评估。这也是为数不多的研究之一，研究了训练人工智能生成以人为本的心理突出内容的新现象。

With the rise of digital art, Uncanny Valley has emerged from an esoteric robotics concept into an infectious memetic phenomenon, with specific memes such as 'Uncanny/Canny Mr. Incredible', or more generally uncanny faces being used as reaction images for humor. Critics and players will now refer to specific media being 'Uncanny' rather than using more general words as 'off-putting', demonstrating uncanniness cementing itself in the public consciousness as examples increasingly abound; ergo digital artists should be aware of evoking the uncanny even with modern rendering technology, as audiences become increasingly discerning of the Uncanny. This is most pertinent in videogames, where rendering is performed in real-time, meaning rendering constraints must be implemented. This potentially confines characters to the Uncanny Valley, as it may not be possible to increase graphical fidelity, thus artists may be left to either accept the uncanny or demaster their work (both undesirable options). This project aims to learn about the Uncanny Valley pertaining to modern skin rendering techniques, using artificial intelligence (specifically GANs) to directly map skin rendering parameters onto user assessments of uncanniness and realism. This can then be reverse engineered to provide automated tools for generatively rendering realistic non-uncanny skin, and predicting audience responses to skin realism, expediting QA testing. The primary experimental stage is to generate a face database with photorealistic skin to be assessed using psychometrics by participants. This is additionally one of few studies looking into the novel phenomena of training AI's to generate human-oriented psychologically salient content.