Men's Hair Styles Approach

Paper #1 - Real-time deep hair matting on mobile devices

Key Notes

• Hand-crafted features for segmentation.
• Employ simple pixel-wise color models to classify hair.
• Fully Convolutional MobileNet Architecture for Hair Segmentation

• HairSegNet
• Pre-trained weights on ImageNet, we dilate all the kernels for the layers with updated resolution by their scale factor
• Upsampling is performed by a simplified version of an inverted MobileNet architecture
• a loss function that promotes perceptually accurate matting output

• HairMatteNet runs twice as fast compared to HairSegNet

Paper #2 - Intuitive, Interactive Beard and Hair Synthesis with Generative Models

Key notes

• Edge detection or image gradients would be an intuitive approach
• Generative adversarial networks (GANs)
• Two-stage pipeline
• First stage focuses on synthesizing realistic facial hair
• Texture synthesis techniques
• pixel-based methods
• stitching-based methods

Generative adversarial networks (GANs) [26] has inspired a large body of high-quality image synthesis and editing approaches

Two Stage Network

• The first stage synthesizes the hair in this region.
• The second stage refines and composites the synthesized hair into the input image.

Close-up images of high-resolution complex structures fail to capture all the complexity of the hair structure, limiting the plausibility of the synthesized images

StyleGAN

Paper - Link

• Generator embeds the input latent code into an intermediate latent space

Deep Convolutional Generative Adversarial Network 

• During training, the generator progressively becomes better at creating images that look real, while the discriminator becomes better at telling them apart. 
• The process reaches equilibrium when the discriminator can no longer distinguish real images from fakes.
• Both the generator and discriminator are defined using the Keras Sequential API.

How to code a Generative Adversarial Network (GAN) in Python

Paper #3 - PROGRESSIVE GROWING OF GANS FOR IMPROVED QUALITY, STABILITY, AND VARIATION

• Our primary contribution is a training methodology for GANs where we start with low-resolution images, and then progressively increase the resolution by adding layers to the networks
• When training the discriminator, we feed in real images that are downscaled to match the current resolution of the network

• MULTI-SCALE STATISTICAL SIMILARITY FOR ASSESSING GAN RESULTS
• Intuitively a small Wasserstein distance indicates that the distribution of the patches is similar

Paper #4 - Training Language GANs from Scratch

ScratchGAN code 

• Generator architecture and reward structure
• Large Batch Sizes for Variance Reduction
• Unlike image GANs, ScratchGAN learns an explicit model of data

Paper #5 - A review of Generative Adversarial Networks (GANs) and its applications in a wide variety of disciplines - From Medical to Remote Sensing

• Generative Adversarial Network (GAN) colab
• Conditional_gan colab
• InterFaceGAN - Interpreting the Latent Space of GANs for Semantic Face Editing
• StyleGAN — Official TensorFlow Implementation
• Hairstyle Transfer — Semantic Editing GAN Latent Code

CelebHair: A New Large-Scale Dataset for Hairstyle Recommendation based on CelebA

More Reads

• Hair Color Digitization through Imaging and Deep Inverse Graphics
• Intuitive, Interactive Beard and Hair Synthesis with Generative Models
• Progressive Color Transfer with Dense Semantic Correspondences
• Local Facial Attribute Transfer through Inpainting
• Generative Single Image Reflection Separation
• ReflectNet - A Generative Adversarial Method for Single Image Reflection Suppression
• A Style-Based Generator Architecture for Generative Adversarial Networks
• K-HAIRSTYLE: A LARGE-SCALE KOREAN HAIRSTYLE DATASET FOR VIRTUAL HAIR EDITING AND HAIRSTYLE CLASSIFICATION
• AttGAN: Facial Attribute Editing by Only Changing What You Want
• HairCLIP: Design Your Hair by Text and Reference Image
• fAshIon after fashion: A Report of AI in Fashion
• Keypoints-Based 2D Virtual Try-on Network System
• StyleCLIP: Text-Driven Manipulation of StyleGAN Imagery
• VOGUE: Try-On by StyleGAN Interpolation Optimization
• Dressing in Order: Recurrent Person Image Generation for Pose Transfer, Virtual Try-on and Outfit Editing

Keep Thinking!!!