Despite the success of diffusion models (DMs), we still lack a thorough understanding of their latent space. While image editing with GANs builds upon latent space, DMs rely on editing the conditions such as text prompts. We present an unsupervised method to discover interpretable editing directions for the latent variables $\mathbf{x}_t \in \mathcal{X}$ of DMs. Our method adopts Riemannian geometry between $\mathcal{X}$ and the intermediate feature maps $\mathcal{H}$ of the U-Nets to provide a deep understanding over the geometrical structure of $\mathcal{X}$. The discovered semantic latent directions mostly yield disentangled attribute changes, and they are globally consistent across different samples. Furthermore, editing in earlier timesteps edits coarse attributes, while ones in later timesteps focus on high-frequency details. We define the curvedness of a line segment between samples to show that $\mathcal{X}$ is a curved manifold. Experiments on different baselines and datasets demonstrate the effectiveness of our method even on Stable Diffusion. Our source code will be publicly available for the future researchers.
3D-aware GANs aim to synthesize realistic 3D scenes such that they can be rendered in arbitrary perspectives to produce images. Although previous methods produce realistic images, they suffer from unstable training or degenerate solutions where the 3D geometry is unnatural. We hypothesize that the 3D geometry is underdetermined due to the insufficient constraint, i.e., being classified as real image to the discriminator is not enough. To solve this problem, we propose to approximate the background as a spherical surface and represent a scene as a union of the foreground placed in the sphere and the thin spherical background. It reduces the degree of freedom in the background field. Accordingly, we modify the volume rendering equation and incorporate dedicated constraints to design a novel 3D-aware GAN framework named BallGAN. BallGAN has multiple advantages as follows. 1) It produces more reasonable 3D geometry; the images of a scene across different viewpoints have better photometric consistency and fidelity than the state-of-the-art methods. 2) The training becomes much more stable. 3) The foreground can be separately rendered on top of different arbitrary backgrounds.
Existing techniques for image-to-image translation commonly have suffered from two critical problems: heavy reliance on per-sample domain annotation and/or inability of handling multiple attributes per image. Recent methods adopt clustering approaches to easily provide per-sample annotations in an unsupervised manner. However, they cannot account for the real-world setting; one sample may have multiple attributes. In addition, the semantics of the clusters are not easily coupled to human understanding. To overcome these, we present a LANguage-driven Image-to-image Translation model, dubbed LANIT. We leverage easy-to-obtain candidate domain annotations given in texts for a dataset and jointly optimize them during training. The target style is specified by aggregating multi-domain style vectors according to the multi-hot domain assignments. As the initial candidate domain texts might be inaccurate, we set the candidate domain texts to be learnable and jointly fine-tune them during training. Furthermore, we introduce a slack domain to cover samples that are not covered by the candidate domains. Experiments on several standard benchmarks demonstrate that LANIT achieves comparable or superior performance to the existing model.
Foreground-aware image synthesis aims to generate images as well as their foreground masks. A common approach is to formulate an image as an masked blending of a foreground image and a background image. It is a challenging problem because it is prone to reach the trivial solution where either image overwhelms the other, i.e., the masks become completely full or empty, and the foreground and background are not meaningfully separated. We present FurryGAN with three key components: 1) imposing both the foreground image and the composite image to be realistic, 2) designing a mask as a combination of coarse and fine masks, and 3) guiding the generator by an auxiliary mask predictor in the discriminator. Our method produces realistic images with remarkably detailed alpha masks which cover hair, fur, and whiskers in a fully unsupervised manner.
In generative adversarial networks, improving discriminators is one of the key components for generation performance. As image classifiers are biased toward texture and debiasing improves accuracy, we investigate 1) if the discriminators are biased, and 2) if debiasing the discriminators will improve generation performance. Indeed, we find empirical evidence that the discriminators are sensitive to the style (\e.g., texture and color) of images. As a remedy, we propose feature statistics mixing regularization (FSMR) that encourages the discriminator's prediction to be invariant to the styles of input images. Specifically, we generate a mixed feature of an original and a reference image in the discriminator's feature space and we apply regularization so that the prediction for the mixed feature is consistent with the prediction for the original image. We conduct extensive experiments to demonstrate that our regularization leads to reduced sensitivity to style and consistently improves the performance of various GAN architectures on nine datasets. In addition, adding FSMR to recently-proposed augmentation-based GAN methods further improves image quality. Code will be publicly available online for the research community.
Generative adversarial networks (GANs) synthesize realistic images from random latent vectors. Although manipulating the latent vectors controls the synthesized outputs, editing real images with GANs suffers from i) time-consuming optimization for projecting real images to the latent vectors, ii) or inaccurate embedding through an encoder. We propose StyleMapGAN: the intermediate latent space has spatial dimensions, and a spatially variant modulation replaces AdaIN. It makes the embedding through an encoder more accurate than existing optimization-based methods while maintaining the properties of GANs. Experimental results demonstrate that our method significantly outperforms state-of-the-art models in various image manipulation tasks such as local editing and image interpolation. Last but not least, conventional editing methods on GANs are still valid on our StyleMapGAN. Source code is available at https://github.com/naver-ai/StyleMapGAN.
Training GANs on videos is even more sophisticated than on images because videos have a distinguished dimension: time. While recent methods designed a dedicated architecture considering time, generated videos are still far from indistinguishable from real videos. In this paper, we introduce ArrowGAN framework, where the discriminators learns to classify arrow of time as an auxiliary task and the generators tries to synthesize forward-running videos. We argue that the auxiliary task should be carefully chosen regarding the target domain. In addition, we explore categorical ArrowGAN with recent techniques in conditional image generation upon ArrowGAN framework, achieving the state-of-the-art performance on categorical video generation. Our extensive experiments validate the effectiveness of arrow of time as a self-supervisory task, and demonstrate that all our components of categorical ArrowGAN lead to the improvement regarding video inception score and Frechet video distance on three datasets: Weizmann, UCFsports, and UCF-101.
Weakly supervised object localization (WSOL) aims to localize the target object using only the image-level supervision. Recent methods encourage the model to activate feature maps over the entire object by dropping the most discriminative parts. However, they are likely to induce excessive extension to the backgrounds which leads to over-estimated localization. In this paper, we consider the background as an important cue that guides the feature activation to cover the sophisticated object region and propose contrastive attention loss. The loss promotes similarity between foreground and its dropped version, and, dissimilarity between the dropped version and background. Furthermore, we propose foreground consistency loss that penalizes earlier layers producing noisy attention regarding the later layer as a reference to provide them with a sense of backgroundness. It guides the early layers to activate on objects rather than locally distinctive backgrounds so that their attentions to be similar to the later layer. For better optimizing the above losses, we use the non-local attention blocks to replace channel-pooled attention leading to enhanced attention maps considering the spatial similarity. Last but not least, we propose to drop background regions in addition to the most discriminative region. Our method achieves state-of-theart performance on CUB-200-2011 and ImageNet benchmark datasets regarding top-1 localization accuracy and MaxBoxAccV2, and we provide detailed analysis on our individual components. The code will be publicly available online for reproducibility.
Normalization techniques, such as batch normalization (BN), have led to significant improvements in deep neural network performances. Prior studies have analyzed the benefits of the resulting scale invariance of the weights for the gradient descent (GD) optimizers: it leads to a stabilized training due to the auto-tuning of step sizes. However, we show that, combined with the momentum-based algorithms, the scale invariance tends to induce an excessive growth of the weight norms. This in turn overly suppresses the effective step sizes during training, potentially leading to sub-optimal performances in deep neural networks. We analyze this phenomenon both theoretically and empirically. We propose a simple and effective solution: at each iteration of momentum-based GD optimizers (e.g. SGD or Adam) applied on scale-invariant weights (e.g. Conv weights preceding a BN layer), we remove the radial component (i.e. parallel to the weight vector) from the update vector. Intuitively, this operation prevents the unnecessary update along the radial direction that only increases the weight norm without contributing to the loss minimization. We verify that the modified optimizers SGDP and AdamP successfully regularize the norm growth and improve the performance of a broad set of models. Our experiments cover tasks including image classification and retrieval, object detection, robustness benchmarks, and audio classification. Source code is available at https://github.com/clovaai/AdamP.
Every recent image-to-image translation model uses either image-level (i.e. input-output pairs) or set-level (i.e. domain labels) supervision at minimum. However, even the set-level supervision can be a serious bottleneck for data collection in practice. In this paper, we tackle image-to-image translation in a fully unsupervised setting, i.e., neither paired images nor domain labels. To this end, we propose the truly unsupervised image-to-image translation method (TUNIT) that simultaneously learns to separate image domains via an information-theoretic approach and generate corresponding images using the estimated domain labels. Experimental results on various datasets show that the proposed method successfully separates domains and translates images across those domains. In addition, our model outperforms existing set-level supervised methods under a semi-supervised setting, where a subset of domain labels is provided. The source code is available at https://github.com/clovaai/tunit