Crypto-Dropout: To Create Unique User-Generated Content Using Crypto Information in Metaverse
Haihan Duan, Xiao Wu, Wei Cai · 2022 IEEE 24th International Workshop on Multimedia Signal Processing (MMSP) · 2022
In a blockchain-driven metaverse, user-generated content (UGC) is the core power for building the metaverse, so an easy-to-use UGC editor is imperative. Specifically, using artificial intelligence (AI) to simplify the UGC creation procedure is promising, e.g., generating images from sketches using generative adversarial networks (GANs). However, the simplicity of these UGC creation methods would lead to weak distinctions between the generated UGC, since the users' created drafts may be very similar. In this paper, we propose Crypto-dropout, a specially designed dropout used in the generative neural networks, which could cause pseudo-random disturbance based on the hash value of user information to generate unique results. With a pilot study, the experimental results demonstrate that the participants have different preferences for the generated images when setting Crypto-dropout in the different layers. Accordingly, we implement a practical profile pictures (PFPs) creation prototype. The proposed Crypto-dropout can provide a novel and general insight for creating unique UGC using generative neural networks.