Explaining in Style: Training a GAN to explain a classifier in StyleSpace

By Oran Lang, Yossi Gandelsman, Michal Yarom, Yoav Wald, Gal Elidan, Avinatan Hassidim, William T. Freeman, Phillip Isola, Amir Globerson, Michal Irani and 1 more. 2021

184 citations recorded by the source corpus.

Image classification models can depend on multiple different semantic attributes of the image. An explanation of the decision of the classifier needs to both discover and visualize these properties. Here we present StylEx, a method for doing this, by training a generative model to specifically explain multiple attributes that underlie classifier decisions. A natural source for such attributes is the StyleSpace of StyleGAN, which is known to generate semantically meaningful dimensions in the image. However, because standard GAN training is not dependent on the classifier, it may not represent those attributes which are important for the classifier decision, and the dimensions of StyleSpace may represent irrelevant at-tributes. To overcome this, we propose a training procedure for a StyleGAN, which incorporates the classifier model, in order to learn a classifier-specific StyleSpace. Explanatory attributes are then selected from this space. These can be used to visualize the effect of changing multiple attributes per image, thus providing image-specific explanations. We apply StylEx to multiple domains, including animals, leaves, faces and retinal images. For these, we show how an image can be modified in different ways to change its classifier output. Our results show that the method finds attributes that align well with semantic ones, generate meaningful image-specific explanations, and are human-interpretable as measured in user-studies.1

Subfields: Artificial Intelligence, Biophysics, Computer Vision and Pattern Recognition, Generative Models, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Rinon Gal
2. Daniel Cohen-Or
3. Bolei Zhou
4. Yotam Nitzan
5. Ming-Hsuan Yang
6. Jing-Hao Xue
7. Ofir Brenner
8. Yulun Zhang

In Biophysics

1. Daniel Cohen-Or
2. Yotam Nitzan
3. Rinon Gal
4. Ofir Brenner
5. Bolei Zhou
6. Ming-Hsuan Yang
7. Jing-Hao Xue
8. Yulun Zhang

In Computer Vision and Pattern Recognition

1. Daniel Cohen-Or
2. Ming-Hsuan Yang
3. Yujiu Yang
4. Weihao Xia
5. Luc Van Gool
6. Jing-Hao Xue
7. Yulun Zhang
8. Bolei Zhou

In Generative Models

1. Daniel Cohen-Or
2. Yujiu Yang
3. Weihao Xia
4. Rinon Gal
5. Yotam Nitzan
6. Jing-Hao Xue
7. Ofir Brenner
8. Yulun Zhang

In Safety, Robustness & Interpretability

1. Daniel Cohen-Or
2. Bolei Zhou
3. Ming-Hsuan Yang

Citations in the corpus

• GAN Inversion: A Survey (2021, methodology): “Some methods [31], [32], [33], [34] use additional encoder networks to learn the inverse mapping of GANs, but their goals are to jointly train the encoder with both the generator and the discriminator, instead of using a trained GAN model .”
• LARGE: Latent-Based Regression through GAN Semantics (2021, methodology): “Furthermore, similar to other methods which rely on StyleGAN [7,29], our method obtains better results when operating on images within the domain used to train the GAN.”
• Imagic: Text-Based Real Image Editing with Diffusion Models (2022, unknown): “Following recent advancements in image synthesis quality [21–24], many works utilized the latent space of pretrained generative adversarial networks (GANs) to perform a variety of image manipulations [3, 15, 30, 36, 49, 50].”