Recent image tone adjustment (or enhancement) approaches have predominantly adopted supervised learning for learning human-centric perceptual assessment. However, these approaches are constrained by intrinsic challenges of supervised learning. Primarily, the requirement for expertly-curated or retouched images escalates the data acquisition expenses. Moreover, their coverage of target styles is confined to stylistic variants inferred from the training data. To surmount the above challenges, we propose an unsupervised learning-based approach for text-based image tone adjustment, CLIPtone, that extends an existing image enhancement method to accommodate natural language descriptions. Specifically, we design a hyper-network to adaptively modulate the pretrained parameters of a back-bone model based on a text description. To assess whether an adjusted image aligns with its text description without a ground-truth image, we utilize CLIP, which is trained on a vast set of language-image pairs and thus encompasses the knowledge of human perception. The major advantages of our approach are threefold: (i) minimal data collection expenses, (ii) support for a range of adjustments, and (iii) the ability to handle novel text descriptions unseen in training. The efficacy of the proposed method is demonstrated through comprehensive experiments including a user study.

Recent image restoration methods can be broadly categorized into two classes: (1) regression methods that recover the rough structure of the original image without synthesizing high-frequency details and (2) generative methods that synthesize perceptually-realistic high-frequency details even though the resulting image deviates from the original structure of the input. While both directions have been extensively studied in isolation, merging their benefits with a single framework has been rarely studied. In this paper, we propose UGPNet, a universal image restoration framework that can effectively achieve the benefits of both approaches by simply adopting a pair of an existing regression model and a generative model. UGPNet first restores the image structure of a degraded input using a regression model and synthesizes a perceptually-realistic image with a generative model on top of the regressed output. UGPNet then combines the regressed output and the synthesized output, resulting in a final result that faithfully reconstructs the structure of the original image in addition to perceptually-realistic textures. Our extensive experiments on deblurring, denoising, and super-resolution demonstrate that UGPNet can successfully exploit both regression and generative methods for high-fidelity image restoration.

Few-shot domain adaptation to multiple domains aims to learn a complex image distribution across multiple domains from a few training images. A naïve solution here is to train a separate model for each domain using few-shot domain adaptation methods. Unfortunately, this approach mandates linearly-scaled computational resources both in memory and computation time and, more importantly, such separate models cannot exploit the shared knowledge between target domains. In this paper, we propose DynaGAN, a novel few-shot domain-adaptation method for multiple target domains. DynaGAN has an adaptation module, which is a hyper-network that dynamically adapts a pretrained GAN model into the multiple target domains. Hence, we can fully exploit the shared knowledge across target domains and avoid the linearly-scaled computational requirements. As it is still computationally challenging to adapt a large-size GAN model, we design our adaptation module to be lightweight using the rank-1 tensor decomposition. Lastly, we propose a contrastive-adaptation loss suitable for multi-domain few-shot adaptation. We validate the effectiveness of our method through extensive qualitative and quantitative evaluations.

For successful semantic editing of real images, it is critical for a GAN inversion method to find an in-domain latent code that aligns with the domain of a pre-trained GAN model. Unfortunately, such in-domain latent codes can be found only for in-range images that align with the training images of a GAN model. In this paper, we propose BDInvert, a novel GAN inversion approach to semantic editing of out- of-range images that are geometrically unaligned with the training images of a GAN model. To find a latent code that is semantically editable, BDInvert inverts an input out-of-range image into an alternative latent space than the original latent space. We also propose a regularized inversion method to find a solution that supports semantic editing in the alternative space. Our experiments show that BDInvert effectively supports semantic editing of out-of-range images with geometric transformations.

A common way to view a 360° video on a 2D display is to crop and render a part of the video as a normal field-of-view (NFoV) video. While users can enjoy natural-looking NFoV videos using this approach, they need to constantly make manual adjustment of the viewing direction not to miss interesting events in the video. In this paper, we propose an interactive and automatic navigation system for comfortable 360° video playback. Our system finds a virtual camera path that shows the most salient areas through the video, generates a NFoV video based on the path, and plays it in an online manner. A user can interactively change the viewing direction while watching a video, and the system instantly updates the path reflecting the intention of the user. To enable online processing, we design our system consisting of an offline pre-processing step, and an online 360° video navigation step. The pre-processing step computes optical flow and saliency scores for an input video. Based on these, the online video navigation step computes an optimal camera path reflecting user interaction, and plays a NFoV video in an online manner. For improved user experience, we also introduce optical flow-based camera path planning, saliency-aware path update, and adaptive control of the temporal window size. Our experimental results including user studies show that our system provides more pleasant experience of watching 360° videos than existing approaches.