Exploring Unlabeled Faces for Novel Attribute Discovery

Bahng, Hyojin; Chung, Sunghyo; Yoo, Seungjoo; Choo, Jaegul

doi:10.1109/cvpr42600.2020.00586

Cited by 11 publications

(6 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, FUNIT requires the labels for training, while S 3 GAN needs a subset of labeled data for the best performance. Recently, Bahng et al [1] has partially addressed this by adopting the pre-trained classifier for extracting domain information. Unlike the previous methods, we aim to design an image-to-image translation model that can be applied without any supervision such as a pre-trained network or supervision on both the train and the test datasets.…”

Section: Related Workmentioning

confidence: 99%

Rethinking the Truly Unsupervised Image-to-Image Translation

Baek

Choi

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

Section: Related Workmentioning

confidence: 99%

Rethinking the Truly Unsupervised Image-to-Image Translation

Baek

Choi

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Recent deep generative models such as flow-based model [67,42,17], auto-regressive models [59,69,63], VAEs [44,43,55,80], and GANs [25,65,5,26,39,40,13,38] have contributed to improvements in unconditional image generation. The literature on conditional image generation explores related tasks including style transfer [23,79,14,34,19,4,8,61], supervised image translation [36,88,24,48,49,47,53,62,77,82,86], unsupervised image translation [51,87,35,10,16,52,68], and semantic image editing [72,9]. Please refer to Liu et al [50] for more details.…”

Section: Related Workmentioning

confidence: 99%

Stay Positive: Non-Negative Image Synthesis for Augmented Reality

Luo

Yang

Xian

et al. 2022

Preprint

View full text Add to dashboard Cite

In applications such as optical see-through and projector augmented reality, producing images amounts to solving non-negative image generation, where one can only add light to an existing image. Most image generation methods, however, are ill-suited to this problem setting, as they make the assumption that one can assign arbitrary color to each pixel. In fact, naive application of existing methods fails even in simple domains such as MNIST digits, since one cannot create darker pixels by adding light. We know, however, that the human visual system can be fooled by optical illusions involving certain spatial configurations of brightness and contrast. Our key insight is that one can leverage this behavior to produce high quality images with negligible artifacts. For example, we can create the illusion of darker patches by brightening surrounding pixels. We propose a novel optimization procedure to produce images that satisfy both semantic and non-negativity constraints. Our approach can incorporate existing state-of-the-art methods, and exhibits strong performance in a variety of tasks including image-to-image translation and style transfer.

show abstract

“…Considerable progress has been made in facial attribute manipulation [18][19][20][21][22][23]. Most methods of facial attribute manipulation are based on generative models.…”

Section: Facial Attribute Manipulationmentioning

confidence: 99%

Mask-aware photorealistic facial attribute manipulation

et al. 2021

View full text Add to dashboard Cite

The technique of facial attribute manipulation has found increasing application, but it remains challenging to restrict editing of attributes so that a face’s unique details are preserved. In this paper, we introduce our method, which we call a mask-adversarial autoencoder (M-AAE). It combines a variational autoencoder (VAE) and a generative adversarial network (GAN) for photorealistic image generation. We use partial dilated layers to modify a few pixels in the feature maps of an encoder, changing the attribute strength continuously without hindering global information. Our training objectives for the VAE and GAN are reinforced by supervision of face recognition loss and cycle consistency loss, to faithfully preserve facial details. Moreover, we generate facial masks to enforce background consistency, which allows our training to focus on the foreground face rather than the background. Experimental results demonstrate that our method can generate high-quality images with varying attributes, and outperforms existing methods in detail preservation.

show abstract

Exploring Unlabeled Faces for Novel Attribute Discovery

Cited by 11 publications

References 22 publications

Rethinking the Truly Unsupervised Image-to-Image Translation

Rethinking the Truly Unsupervised Image-to-Image Translation

Stay Positive: Non-Negative Image Synthesis for Augmented Reality

Mask-aware photorealistic facial attribute manipulation

Contact Info

Product

Resources

About