DeepNude algorithm “removing” clothes – Welfare Bar http://fulibus.net/deepnude.html
First, the file is as shown
Deep learning computer vision (speculated)
Image Inpainting
You can refer to NVIDIA’s paper using partial convolution and partial convolution-based filling to repair images with irregular holes.
The code part of the paper is converted. Paper address: https://arxiv.org/abs/1804.07723 and https://arxiv.org/abs/1811.11718
In the Image_Inpainting experimental test (https://api.isoyu.com/Deepnude/Image_Inpainting(NVIDIA_2018).mp4) video image interface, you only need to use the tool to simply smear unwanted content in the image. Even if the shape is very irregular, NVIDIA’s model can realistically “restore” the image. The image fills the smeared blank. It can be described as a one-click P image, and there are “no PS traces”. This research is based on the team of Liu from Guilin, NVIDIA. Ji Changxin is also continuously following them, they have released a deep learning method that can edit images or rebuild damaged images, even if the image has a hole or missing pixels. This is currently the most advanced method in the country in 2018.
Pix2Pix (requires paired data)
Reference paper: https://arxiv.org/abs/1611.07004
Below is the output generated after training the Pix2Pix model for 200 epochs.
For more information, you can check https://github.com/tensorflow/docs/blob/master/site/en/r2/tutorials/generative/pix2pix.ipynb
CycleGAN (no paired data required)
CycleGAN uses a cycle consistency loss function to achieve training without paired data. In other words, it can convert from one domain to another without a one-to-one mapping between the source and target domains. This opens up the possibility of performing many interesting tasks, such as photo enhancement, image coloring, style transfer, etc. Only the source and target datasets are needed.
Reference paper: https://arxiv.org/abs/1703.10593
For more information https://github.com/tensorflow/docs/blob/master/site/en/r2/tutorials/generative/cyclegan.ipynb
Windows version of DeepNude usage process
DeepNude can truly achieve the purpose of image-to-image, and the generated images are more realistic.
Note: Delete the color.cp36-win_amd64.pyd file in the deepnude root directory, then add the color.py (https://api.isoyu.com/Deepnude/color.py) file to get the advanced version of deepnude.
Personal Suggestions for Fighting to Re-list DeepNude
1. Size. Including 156M DeepNude_Windows_v2.0.0.zip and 1.90G pyqtlib.rar;
2. Speed. Converting images takes 30 seconds;
3. Content. Automatically remove clothes from women using image-to-image neural networks to reveal their nudity. This application is a misuse of deep learning.
* DeepNude can be implemented using Tensorflow and model compression techniques.
* DeepNude should change its current disrespectful practices towards women.
Summary
In fact, image-to-image is not necessary. We can use GANs to directly generate images from random values or from text.
Obj-GAN: https://github.com/jamesli1618/Obj-GAN
StoryGAN: https://github.com/yitong91/StoryGAN
The new AI technology developed by Microsoft Research AI can understand natural language descriptions, sketch, synthesize images, and then refine details based on the sketch framework and individual words provided in the text. In other words, this network can generate images of the same scene based on textual descriptions that describe everyday scenes.
The current optimal text-to-image generation model can generate realistic bird images based on single-sentence descriptions. However, text-to-image generators go far beyond generating a single image for a single sentence. Given a multi-sentence paragraph, it generates a series of images, each corresponding to a sentence, fully visualizing the entire story.
More https://github.com/yuanxiaosc follow and you will be fine
This article was created by Ji Changxin, article address: https://blog.isoyu.com/archives/deepnude.html
Licensed under Creative Commons Attribution 4.0 International License. Unless otherwise stated, all articles are original or translated by this site, please be sure to indicate the source before reprinting. The last edited time was: June 29, 2019 at 06:19 AM