Abstract:Creating and updating pixel art character sprites with many frames spanning different animations and poses takes time and can quickly become repetitive. However, that can be partially automated to allow artists to focus on more creative tasks. In this work, we concentrate on creating pixel art character sprites in a target pose from images of them facing other three directions. We present a novel approach to character generation by framing the problem as a missing data imputation task. Our proposed generative adversarial networks model receives the images of a character in all available domains and produces the image of the missing pose. We evaluated our approach in the scenarios with one, two, and three missing images, achieving similar or better results to the state-of-the-art when more images are available. We also evaluate the impact of the proposed changes to the base architecture.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the **problem of generating pixel - art character sprites in different poses**. Specifically, the author proposes a new method based on Generative Adversarial Networks (GAN) to infer the missing pose images from the existing character images. This problem is framed as a **missing - data imputation task**, that is, using all the other available pose images to generate the missing pose images. #### Problem Background 1. **Repetitive Work in Character Design**: - In the game development process, creating and updating pixel - art character sprites with multiple animation frames and poses is a time - consuming and easily repetitive task. - Although character design requires a high degree of creativity, some of these tasks can be automated so that artists can focus on more creative tasks. 2. **Limitations of Existing Methods**: - Some existing works use techniques such as Variational Auto - Encoders (VAE), Generative Adversarial Networks (GAN) and Convolutional Neural Networks (CNN) to generate character images, but these methods usually only handle image conversion between two domains (for example, from front to side) and do not fully utilize all the available information. #### Main Contributions of the Paper 1. **Multi - pose Generation Model**: - A GAN model based on the CollaGAN architecture is proposed, which can generate multiple character poses through a single generator/discriminator. 2. **Improving Generation Quality by Using More Information**: - Experiments show that using more available information can significantly improve the quality of the generated sprite images. 3. **Improving the CollaGAN Architecture**: - The CollaGAN architecture is modified to enhance the quality of the generated images, and the effects of these modifications are demonstrated through ablation experiments. #### Method Overview - **Generator**: Receives character images from multiple source domains and generates the missing pose images in the target domain. - **Discriminator**: Distinguishes between the generated images and the real images and classifies the domain to which the image belongs. - **Loss Function**: Includes regression loss, cycle - consistency loss, structural similarity loss, adversarial loss and domain - classification loss. Through this method, the paper solves the problem of how to efficiently generate high - quality pixel - art character sprites, especially when given multiple poses.

A Missing Data Imputation GAN for Character Sprite Generation

Pixel art character generation as an image-to-image translation problem using GANs

Univariate Time Series missing data Imputation using Pix2Pix GAN

GAGIN: generative adversarial guider imputation network for missing data

Imputation of Missing Data with Class Imbalance using Conditional Generative Adversarial Networks

Font Generation with Missing Impression Labels

Neural Puppet: Generative Layered Cartoon Characters

Full-body High-resolution Anime Generation with Progressive Structure-conditional Generative Adversarial Networks

DPGAN: A Dual-Path Generative Adversarial Network for Missing Data Imputation in Graphs

A novel 8-connected Pixel Identity GAN with Neutrosophic (ECP-IGANN) for missing imputation

Improved generative adversarial imputation networks for missing data

A time series continuous missing values imputation method based on generative adversarial networks

MisGAN: Learning from Incomplete Data with Generative Adversarial Networks

Recovering from missing data in population imaging – Cardiac MR image imputation via conditional generative adversarial nets

A data imputation method for multivariate time series based on generative adversarial network.

Towards the Automatic Anime Characters Creation with Generative Adversarial Networks

Augmenting Character Designers Creativity Using Generative Adversarial Networks

Missing data imputation with adversarially-trained graph convolutional networks

Multivariate Time Series Missing Value Filling Based on Trans-GAN Model

Deep Generative Imputation Model for Missing Not At Random Data

Missing Features Reconstruction Using a Wasserstein Generative Adversarial Imputation Network