Abstract:In recent years, with the rapid development of artificial intelligence, image generation based on deep learning has dramatically advanced. Image generation based on Generative Adversarial Networks (GANs) is a promising study. However, since convolutions are limited by spatial-agnostic and channel-specific, features extracted by traditional GANs based on convolution are constrained. Therefore, GANs are unable to capture any more details per image. On the other hand, straightforwardly stacking of convolutions causes too many parameters and layers in GANs, which will lead to a high risk of overfitting. To overcome the aforementioned limitations, in this paper, we propose a new GANs called Involution Generative Adversarial Networks (GIU-GANs). GIU-GANs leverages a brand new module called the Global Information Utilization (GIU) module, which integrates Squeeze-and-Excitation Networks (SENet) and involution to focus on global information by channel attention mechanism, leading to a higher quality of generated images. Meanwhile, Batch Normalization(BN) inevitably ignores the representation differences among noise sampled by the generator, and thus degrade the generated image quality. Thus we introduce Representative Batch Normalization(RBN) to the GANs architecture for this issue. The CIFAR-10 and CelebA datasets are employed to demonstrate the effectiveness of our proposed model. A large number of experiments prove that our model achieves state-of-the-art competitive performance.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve lies in the limitations of existing convolutional neural network (CNN) - based generative adversarial networks (GANs) in image generation tasks. Specifically, these problems include: 1. **Spatial - independence and channel - specificity**: Traditional convolution operations are restricted by spatial - independence and channel - specificity, making it difficult for the model to capture global information and in - depth details in the image. 2. **Excessive parameters and over - fitting risk**: Directly stacking convolutional layers will lead to too many parameters and layers in GANs, increasing the risk of over - fitting, thus affecting the quality of the generated image. 3. **Problems with batch normalization (BN)**: BN will ignore the representation differences between the generator - sampled noises, thereby reducing the quality of the generated image. To solve the above problems, the author proposes GIU - GANs (Global Information Utilization for GANs) and introduces a new module - the GIU module. The GIU module combines the Squeeze - and - Excitation (SE) module and the involution operation, focuses on global information through the channel - attention mechanism, and enhances the quality of the generated image. In addition, the author also introduces representative batch normalization (RBN) to improve BN in the generator, in order to better handle the representation differences of different noises and further improve the quality of the generated image. ### Summary of main contributions: - **Designed a new GIU module**: This module uses global information to improve the GAN's feature extraction ability, thereby enhancing the quality of the generated image. - **Introduced RBN**: Makes GAN pay more attention to the expression of representative features and improves the quality of the generated image. - **Adopted two techniques**: Spectral normalization and WGAN - GP, to stabilize GAN training and improve the quality of the generated image. These improvements make GIU - GANs perform better than many classic GAN models on the CIFAR - 10 and CelebA datasets.

GIU-GANs: Global Information Utilization for Generative Adversarial Networks

SpatialGAN: Progressive Image Generation Based on Spatial Recursive Adversarial Expansion

UGC: Unified GAN Compression for Efficient Image-to-Image Translation

Incremental Focal Loss GANs.

SUGAN: A Stable U-Net Based Generative Adversarial Network

A Novel Generator With Auxiliary Branch for Improving GAN Performance

GL-GAN: Adaptive Global and Local Bilevel Optimization model of Image Generation

GIQA: Generated Image Quality Assessment

Two Birds with One Stone: Iteratively Learn Facial Attributes with GANs.

Image generation step by step: animation generation-image translation

xAI-GAN: Enhancing Generative Adversarial Networks via Explainable AI Systems

Generalized Visual Quality Assessment of GAN-Generated Face Images

Improving Global Adversarial Robustness Generalization With Adversarially Trained GAN

Exploring conditional pixel-independent generation in GAN inversion for image processing

CBAM-GAN: Generative Adversarial Networks Based on Convolutional Block Attention Module

Improving the Speed and Quality of GAN by Adversarial Training

Towards the Gradient Vanishing, Divergence Mismatching and Mode Collapse of Generative Adversarial Nets

Stacked Siamese Generative Adversarial Nets: A Novel Way to Enlarge Image Dataset

Stabilizing and Improving Training of Generative Adversarial Networks Through Identity Blocks and Modified Loss Function

A Neuro-AI Interface for Evaluating Generative Adversarial Networks