Rethinking and Improving the Robustness of Image Style Transfer
By Pei Wang, Yijun Li, Nuno Vasconcelos. 2021
108 citations recorded by the source corpus.
Extensive research in neural style transfer methods has shown that the correlation between features extracted by a pre-trained VGG network has a remarkable ability to capture the visual style of an image. Surprisingly, however, this stylization quality is not robust and often degrades significantly when applied to features from more advanced and lightweight networks, such as those in the ResNet family. By performing extensive experiments with different network architectures, we find that residual connections, which represent the main architectural difference between VGG and ResNet, produce feature maps of small entropy, which are not suitable for style transfer. To improve the robustness of the ResNet architecture, we then propose a simple yet effective solution based on a softmax transformation of the feature activations that enhances their entropy. Experimental results demonstrate that this small magic can greatly improve the quality of stylization results, even for networks with random weights. This suggests that the architecture used for feature extraction is more important than the use of learned weights for the task of style transfer.
Subfields: Computer Vision and Pattern Recognition, Generative Models
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Vision and Pattern Recognition
- Yujun Shen
- Yinghao Xu
- Jiapeng Zhu
- Bolei Zhou
- Ceyuan Yang
- Daniel Cohen-Or
- Eli Shechtman
- Chen Change Loy
In Generative Models
Citations in the corpus
- Learning to Imagine: Diversify Memory for Incremental Learning using Unlabeled Data (2022, methodology): “The gram matrix is adopted on the level of feature map to encode such semantically-irrelevant information following prior works [8, 13, 23, 24, 31].”
- One-Shot Generative Domain Adaptation (2021, background): “Here, we qualitatively compare our approach with the state-of-the-art method, SWAG [12] regarding the style transfer task.”
- U-KAN Makes Strong Backbone for Medical Image Segmentation and Generation (2024, background): “Unlike other categories of generative models [7, 17, 35, 36, 61, 83], such as Variational Autoencoders (VAE) [36], Generative Adversarial Networks (GANs) [7, 17, 35, 93], and vector quantization methods [15, 82], diffusion models introduce a novel generative paradigm.”