Fig 1. Advancements of GANs from StyleGAN — [6]

Introduction

When considering the literature of Generative Adversarial Networks, Wasserstein GANs have become one of the key concepts due to their training stability compared to conventional GANs. In this article, I will be going through the concept of gradient-penalty-based WGAN.

The article is organized as follows,

  1. Comparison between GAN and WGAN
  2. The mathematical background of Gradient-Penalty based WGAN
  3. Implementation of WGAN on Celeba-Face dataset from scratch using PyTorch
  4. Discussion of the results

If you are new to Generative Adversarial Networks, please check my previous articles on,

  1. Image-to-Image Translation Using…


Fig 1. Object Detections on different platforms

Introduction

With the general availability of capable processors, machine learning frameworks, and the advent of versatile machine learning networks, AI at the edge — i.e. AI-IoT — has become a practical reality. The chief bottleneck in deploying a real-world AI-based application on an IoT device would be the inference speed of the selected network (as measured through FPS) on the given hardware.

It goes without saying that throwing more hardware at the problem is probably an easy though expensive solution. But what can one do to gain the best performance given a fixed low-cost hardware platform? What are the simple knobs…


ScholarX is an exclusive 6-month program aimed at providing mentoring support to a selected pool of high Potential undergraduate students based in Sri Lanka ideally by a Sri Lankan expat currently engaged with one of the world’s top universities or Fortune 500 companies. It’s our free premium mentoring platform by Sri Lankans for Sri Lankans working towards creating a culture of knowledge and expertise sharing without the limitation of geographical borders.


Fig 1: From the paper “Image-to-Image Translation with Conditional Adversarial Networks”¹

Introduction

Image-to-Image translation is one of the most exciting areas in the fields of image processing, machine vision. There are different deep learning architectures/ loss functions that have been implemented by targeting different specific Image-to-Image translation tasks such as,

  1. Color synthesizing from Edges
  2. Grey Scale to Colored conversion
  3. Season Translation
  4. Motion Transfer
  5. Deep Fake Generation

But the main drawback of these methods is, they are capable of targeting only a single and specific task. This is where the Generative Adversarial Networks comes into play.

In the paper “Image-to-Image Translation with Conditional Adversarial Networks”¹, The authors have proposed a General-Purpose…


Fig 1: DCGAN for MNIST

What is Deep Convolutional Generative Adversarial Network?

Deep Convolutional Generative Adversarial Networks or DCGAN was a state-of-the-art model released with the paper called “Unsupervised Representation Learning with Deep Convolutional Adversarial Networks¹” in 2016. The concept of GAN has become much important concept because this gives the way to the Generative aspects of Deep Learning. Here I am presenting my experience on DCGAN.

How does it work?

DCGAN is basically a GAN (Generative Adversarial Net) architecture that using Convolutions. When considering the GANs, there are 2 main networks called Discriminator and the Generator which are trying to improve each other.

Model Architecture

Let’s get into deep in Generator and the Discriminator.

Generator

The generator…

Udith Haputhanthri

Bio-Medical Engineering Undergraduate | University of Moratuwa | AI enthusiast | Music Lover

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store