Rethinking Image Skip Connections in StyleGAN2
Not on the map. Why?
Abstract (arXiv)
Various models based on StyleGAN have gained significant traction in the field of image synthesis, attributed to their robust training stability and superior performances. Within the StyleGAN framework, the adoption of image skip connection is favored over the traditional residual connection. However, this preference is just based on empirical observations; there has not been any in-depth mathematical analysis on it yet. To rectify this situation, this brief aims to elucidate the mathematical meaning of the image skip connection and introduce a groundbreaking methodology, termed the image squeeze connection, which significantly improves the quality of image synthesis. Specifically, we analyze the image skip connection technique to reveal its problem and introduce the proposed method which not only effectively boosts the GAN performance but also reduces the required number of network parameters. Extensive experiments on various datasets demonstrate that the proposed method consistently enhances the performance of state-of-the-art models based on StyleGAN. We believe that our findings represent a vital advancement in the field of image synthesis, suggesting a novel direction for future research and applications.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 48 of its 60 references that are in the map. First posted on arXiv 8 Jul 2024.
Builds on
48 of 60 references.
References · 48
- 2015Deep Residual Learning for Image Recognitioncited by 51,809
- 2020Denoising Diffusion Probabilistic Modelscited by 20,851
- 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
- 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 8,964
- 2018A Style-Based Generator Architecture for Generative Adversarial Networkscited by 5,986
- 2015Rethinking the Inception Architecture for Computer Visioncited by 5,451
- 2016Image-to-Image Translation with Conditional Adversarial Networkscited by 4,510
- 2019Analyzing and Improving the Image Quality of StyleGANcited by 3,580
- 2017Progressive Growing of GANs for Improved Quality, Stability, and Variationcited by 2,967
- 2016Improved Techniques for Training GANscited by 2,938
- 2018Large Scale GAN Training for High Fidelity Natural Image Synthesiscited by 2,609
- 2017Improved Training of Wasserstein GANscited by 2,223
36 more
- 2017Arbitrary Style Transfer in Real-time with Adaptive Instance Normalizationcited by 1,998
- 2018Spectral Normalization for Generative Adversarial Networkscited by 1,618
- 2015LSUN: Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loopcited by 1,350
- 2020Training Generative Adversarial Networks with Limited Datacited by 1,184
- 2021Alias-Free Generative Adversarial Networkscited by 1,181
- 2019StarGAN v2: Diverse Image Synthesis for Multiple Domainscited by 1,095
- 2017StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translationcited by 1,036
- 2016Least Squares Generative Adversarial Networkscited by 1,000
- 2019Improved Precision and Recall Metric for Assessing Generative Modelscited by 925
- 2018Self-Attention Generative Adversarial Networkscited by 809
- 2018Which Training Methods for GANs do actually Converge?cited by 672
- 2017Wasserstein GANcited by 635
- 2016StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networkscited by 575
- 2022StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasetscited by 489
- 2018Assessing Generative Models via Precision and Recallcited by 401
- 2020Differentiable Augmentation for Data-Efficient GAN Trainingcited by 361
- 2018cGANs with Projection Discriminatorcited by 253
- 2017StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networkscited by 229
- 2023StyleGAN-T: Unlocking the Power of GANs for Fast Large-Scale Text-to-Image Synthesiscited by 212
- 2019Consistency Regularization for Generative Adversarial Networkscited by 134
- 2022Wavelet Diffusion Models are fast and scalable Image Generatorscited by 116
- 2023DiffiT: Diffusion Vision Transformers for Image Generationcited by 104
- 2019AutoGAN: Neural Architecture Search for Generative Adversarial Networkscited by 94
- 2021Unleashing Transformers: Parallel Token Prediction with Discrete Absorbing Diffusion for Fast High-Resolution Image Generation from Vector-Quantized Codescited by 71
- 2018Improving the Improved Training of Wasserstein GANs: A Consistency Term and Its Dual Effectcited by 49
- 2023GETAvatar: Generative Textured Meshes for Animatable Human Avatarscited by 27
- 2019PEPSI++: Fast and Lightweight Network for Image Inpaintingcited by 24
- 2022Polarity Sampling: Quality and Diversity Control of Pre-Trained Generative Networks via Singular Valuescited by 21
- 2023Diverse Inpainting and Editing with GAN Inversioncited by 20
- 2019Generalization in Generative Adversarial Networks: A Novel Perspective from Privacy Protectioncited by 19
- 2023StyleInV: A Temporal Style Modulated Inversion Network for Unconditional Video Generationcited by 19
- 2021Feature Statistics Mixing Regularization for Generative Adversarial Networkscited by 9
- 2022Generator Knows What Discriminator Should Learn in Unconditional GANscited by 7
- 2021Generative Convolution Layer for Image Generationcited by 5
- 2021A Novel Generator with Auxiliary Branch for Improving GAN Performancecited by 4
- 2022GLeaD: Improving GANs with A Generator-Leading Taskcited by 4
Built on by
Citing papers per year: 2025: 1.
1 paper.