DeepNude's algorithm and general image generation theory and practice research, including pix2pix, CycleGAN, UGATIT, DCGAN, SinGAN, ALAE, mGANprior, StarGAN-v2 and VAE models (TensorFlow2 implementation). DeepNude的算法以及通用生成对抗网络(GAN,Generative Adversarial Network)图像生成的理论与实践研究。
This repository contains the pix2pixHD algorithms(proposed by NVIDIA) of DeepNude, and more importantly, the general image generation theory and practice behind DeepNude.
This resource includes the TensorFlow2 (Pytorch | PaddlePaddle) implementation of image generation models such as pix2pix, CycleGAN, UGATIT, DCGAN, SinGAN, VAE, ALAE, mGANprior and StarGAN-v2, which can be used to systematically learn to Generating Adversarial Network (GAN).
This repository is being sponsored by the following tool; please help to support us by taking a look and signing up to a free trial.
DeepNude uses a slightly modified version of the pix2pixHD GAN architecture, quoted from deepnude_official. pix2pixHD is a general-purpose Image2Image technology proposed by NVIDIA. Obviously, DeepNude is the wrong application of artificial intelligence technology, but it uses Image2Image technology for researchers and developers working in other fields such as fashion, film and visual effects.
This section provides a fake image generation demo that you can use as you wish. They are fake images generated by StyleGAN without any copyright issues. Note: Each time you refresh the page, a new fake image will be generated, pay attention to save!
This section provides a demo of Image-to-Image Demo: Black and white stick figures to colorful faces, cats, shoes, handbags. DeepNude software mainly uses Image-to-Image technology, which theoretically converts the images you enter into any image you want. You can experience Image-to-Image technology in your browser by clicking Image-to-Image Demo below.
An example of using this demo is as follows:
In the left side box, draw a cat as you imagine, and then click the process button, you can output a model generated cat.
DeepNude is a pornographic software that is forbidden by minors. If you are not interested in DeepNude itself, you can skip this section and see the general Image-to-Image theory and practice in the following chapters.
DeepNude_software_itself content:
Recognition and conversion of five types of images [porn, hentai, sexy, natural, drawings]. Correct application of image-to-image technology.
NSFW(Not Safe/Suitable For Work) is a large-scale image dataset containing five categories of images [porn, hentai, sexy, natural, drawings]. Here, CycleGAN is used to convert different types of images, such as porn->natural.
This section describes DeepNude-related AI/Deep Learning theory (especially computer vision) research. If you like to read the paper and use the latest papers, enjoy it.
Result
Image-to-Image Translation with Conditional Adversarial Networks is a general solution for the use of conditional confrontation networks as an image-to-image conversion problem proposed by the University of Berkeley.
DeepNude mainly uses this Image-to-Image(Pix2PixHD) technology.
Result
Get high resolution images from the semantic map. The semantic graph is a color picture. The different color blocks on the map represent different kinds of objects, such as pedestrians, cars, traffic signs, buildings, and so on. Pix2PixHD takes a semantic map as input and produces a high-resolution, realistic image. Most of the previous techniques can only produce rough, low-resolution images that don't look real. This research has produced images with a resolution of 2k by 1k, which is very close to full HD photos.
Result
CycleGAN uses a cycle consistency loss to enable training without the need for paired data. In other words, it can translate from one domain to another without a one-to-one mapping between the source and target domain. This opens up the possibility to do a lot of interesting tasks like photo-enhancement, image colorization, style transfer, etc. All you need is the source and the target dataset.
Result
UGATIT is a novel method for unsupervised image-to-image translation, which incorporates a new attention module and a new learnable normalization function in an end-to-end manner. UGATIT can do both image conversions that require Holistic Changes, and image conversions that require Large Shape Changes. It can be seen as an enhanced version of CycleGAN, a more efficient general image conversion framework.
Result
Source A + Source B (Style) = ?
StyleGAN can not only generate fake images source A and source B, but also combine the content of source A and source B from different strengths, as shown in the following table.
Style level (from source b) | Source A | Source B |
---|---|---|
High level (coarse) | all colors (eyes, hair, light) and details facial features from Source A | inherit advanced facial features from Source B, such as posture, general hair style, facial shape and glasses |
Medium level | posture, general facial shape and glasses come from source a | inherits the middle level facial features of source B, such as hair style, open / closed eyes |
High level (fine) | the main facial content comes from source a | inherits the advanced facial features of source B, such as color scheme and microstructure |
Without increasing the amount of calculation of StyleGAN, while solving the image artifacts generated by StyleGAN and obtaining high-quality images with better details, StyleGAN2 implements a new SOTA for unconditional image modeling tasks.
Result
In the image interface of Image_Inpainting(NVIDIA_2018).mp4 video, you only need to use tools to simply smear the unwanted content in the image. Even if the shape is very irregular, NVIDIA's model can “restore” the image with very realistic The picture fills the smeared blank. It can be described as a one-click P picture, and "no ps traces." The study was based on a team from Nvidia's Guilin Liu et al. who published a deep learning method that could edit images or reconstruct corrupted images, even if the images were worn or lost pixels. This is the current 2018 state-of-the-art approach.
ICCV2019 Best paper - Marr prize
Result
We introduce SinGAN, an unconditional generative model that can be learned from a single natural image. Our model is trained to capture the internal distribution of patches within the image, and is then able to generate high quality, diverse samples that carry the same visual content as the image. SinGAN contains a pyramid of fully convolutional GANs, each responsible for learning the patch distribution at a different scale of the image. This allows generating new samples of arbitrary size and aspect ratio, that have significant variability, yet maintain both the global structure and the fine textures of the training image. In contrast to previous single image GAN schemes, our approach is not limited to texture images, and is not conditional (i.e. it generates samples from noise). User studies confirm that the generated samples are commonly confused to be real images. We illustrate the utility of SinGAN in a wide range of image manipulation tasks.
Result
Although studied extensively, the issues of whether they have the same generative power of GANs, or learn disentangled representations, have not been fully addressed. We introduce an autoencoder that tackles these issues jointly, which we call Adversarial Latent Autoencoder (ALAE). It is a general architecture that can leverage recent improvements on GAN training procedures.
Result
Despite the success of Generative Adversarial Networks (GANs) in image synthesis, applying trained GAN models to real image processing remains challenging. Previous methods typically invert a target image back to the latent space either by back-propagation or by learning an additional encoder. However, the reconstructions from both of the methods are far from ideal. In this work, we propose a novel approach, called mGANprior, to incorporate the well-trained GANs as effective prior to a variety of image processing tasks.
Result
A good image-to-image translation model should learn a mapping between different visual domains while satisfying the following properties: 1) diversity of generated images and 2) scalability over multiple domains. Existing methods address either of the issues, having limited diversity or multiple models for all domains. We propose StarGAN v2, a single framework that tackles both and shows significantly improved results over the baselines.
Result
Recent deep image-to-image translation techniques allow fast generation of face images from freehand sketches. However, existing solutions tend to overfit to sketches, thus requiring professional sketches or even edge maps as input. To address this issue, our key idea is to implicitly model the shape space of plausible face images and synthesize a face image in this space to approximate an input sketch.
These models are based on the latest implementation of TensorFlow2.
This section explains DeepNude-related AI/Deep Learning (especially computer vision) code practices, and if you like to experiment, enjoy them.
Use the Pix2Pix model (Conditional Adversarial Networks) to implement black and white stick figures to color graphics, flat houses to stereoscopic houses and aerial maps to maps.
Under development... First you can use the official implementation
The CycleGAN neural network model is used to realize the four functions of photo style conversion, photo effect enhancement, landscape season change, and object conversion.
DCGAN is used to achieve random number to image generation tasks, such as face generation.
VAE is used to achieve random number to image generation tasks, such as face generation.
Use VGG19 to achieve image style migration effects, such as photo changes to oil paintings and comics.
..........................................................................
If you are a user of PaddlePaddle, you can refer to the paddlepaddle version of the above model image generation model library paddegan.
DeepFakes can be seen as an upgraded version of DeepNude, which uses a deep learning model to generate a series of techniques that can be faked, such as fake images, fake audio, and fake videos.
Using deep fake machine learning to create a video from an image and a source video. Related paper: First Order Motion Model for Image Animation
Speaker's Video + Any Image = Fake Video
click to try MyVoiceYourFace Online!
One photo + One audio = Composite Video
We propose a temporal GAN capable of producing animated faces using only a still image of a person and an audio clip containing speech. The videos generated using this model do not only produce lip movements that are synchronized with the audio but also exhibit characteristic facial expressions such as blinks, brow raises etc. This extends our previous model by separately dealing with audio-visual synchronization and expression generation. Our improved model works on "in-the-wild" unseen faces and is capable of capturing the emotion of the speaker and reflecting it in the facial expression.
Interested in DeepFakes?
Click to start systematic learning DeepFakes