Object Detection Made Easy with PyTorch Hub — This article is an introductory tutorial where we download the pre-trained YOLOv5 from PyTorch Hub and perform object detection on sample images. YOLO (You Look Only Once) is a well-known name for object detection. Joseph Redmon from the University of Washington developed the original versions (YOLOv1, v2, and v3) and…

Turn a Horse into a Zebra and vice versa with the Magic of Self-Supervised Learning — This article explains how CycleGAN (aka Cycle-consistent GAN) works, which is well-known for the demo that translates horse into zebra and vice versa. In the previous article, we discussed pix2pix which does similar image translations. In fact, the same people that worked on pix2pix developed CycleGAN to overcome problems…

Image-to-image Translation with a Conditional Adversarial Networks — This article explains the mechanism of the pix2pix model, which translates an image to another one with a totally different look and feel. They use the word “translate” because of what it does, which is hard to explain in words, so I’d like you to watch the following video to…

Specify What Images To Generate With 1 Simple Yet Powerful Change — GAN and DCGAN generate random images. As such, we have almost no control over what images to generate. However, CGAN can let us specify a condition so that we can tell it what images to generate. The trick is that we convert label values into feature vectors using a learnable…

A short follow-up on the previous DCGAN (Deep Convolutional Generative Adversarial Network) articles, but for color images. — This is a short follow-up on DCGAN (Deep Convolutional Generative Adversarial Network) applied for color images. We’ll download a pre-trained DCGAN model from the Torch Hub and generate fashion images. The model was introduced in the paper Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. One of the authors…

DCGAN generates MNIST images with dramatically better quality — In this article, we incorporate the idea from DCGAN to improve the simple GAN model that we trained in the previous article. 1. DCGAN — Our Reference Model We refer to PyTorch’s DCGAN tutorial for DCGAN implementation. We are especially interested in the convolutional (Conv2d) layers as we believe they will improve how the discriminator extract…

Simple Implementation with PyTorch — In this article, I’ll explain how GAN (Generative Adversarial Network) works while implementing it step-by-step with PyTorch. GAN is a generative model that produces random images given a random input. We will define the model and train it. 1. Introduction 1.1. Ian Goodfellow and GAN As you may already know, Ian Goodfellow proposed GAN in 2014. I…

How can we compress BERT while keeping 97% of the performance? — In 2019, the team at Hugging Face released a model based on BERT that was 40% smaller and 60% faster while retaining 97% of the language understanding capability. They called it DistilBERT. So, exactly how small is DistilBERT? …

How did RoBERTa outperform XLNet with no architectural changes to the original BERT? — In 2018, Jacob Devlin et al. (Google AI Language) released BERT that achieved state-of-the-art results on nine NLP tasks, inspiring other researchers to come up with further improvements. Soon, newer models superseded the original BERT models. At least, that was how it looked. One such example was XLNet by Zhilin…