Deep Learning Weekly Issue #109

AR shoe fitting from Nike, new filters in Snapchat, 3D pose estimation, hyper parameting tuning in Keras...

Hey folks,

This week in deep learning, we bring you a new Nike app that measures your foot size, a Deepfake of Salvador Dali, a new Amazon Go store in NYC, and Léon Bottou’s thoughts on learning causation.

You may also enjoy a deep dive into Snapchat’s latest filter (probably a GAN?), a review of 3D pose estimation techniques, a new hyperparameter tuning tool for Keras, and millions of segmentation masks added to Open Images.

As always happy reading and hacking. If you have something you think should be in next week's issue, find us on Twitter: @dl_weekly.

Until next week!


Nike wants your sneakers to fit better, so it’s using AR [Wired]

Nike+ members can use the Nike app on their phone to scan their feet and get precise measurements for shoes.

Amazon Go supermarket opens in New York City [Fortune]
Amazon quietly opened its first cashierless, all computer vision, grocery store in Manhattan this week.

Deepfake Salvador Dalí takes selfies with museum visitors [The Verge]
The Dali Museum used the same tech responsible for Deepfakes to create a virtual clone of the late artist to greet and interact with visitors.

Deep learning could reveal why the world works the way it does [Technology Review]

Facebook’s Léon Bottou on how machine learning can move past correlation and start understanding causation.


Implementing SPADE using Fastai

A PyTorch implementation of Nvidia’s paper using GANs to generate photo-realistic images from crude sketches.

Fun with Snapchat’s Gender Swapping Filter

A deep dive into the uncanny new Snapchat filter. It’s probably a GAN?

Keras Tuner: hypertuning for humans

Keras Tuner, a late announcement from Google I/O, is a high level hyperparameter tuner for the framework complete with a hosted visualization tool.

A 2019 guide to 3D Human Pose Estimation

A nice literature review of 3D pose estimation.


Announcing Open Images V5 and the ICCV 2019 Open Images Challenge
Open Images V5 now contain 2.8 million segmentation masks.

Libraries & Code

Snips Tract: A Rust Neural Network library for the Edge

An on-device neural network runtime written in Rust. Great for low power, embedded devices.

[Github] tensorflow/graphics: Computer Graphics Meets Deep Learning

3D graphics capabilities for TensorFlow.

[Github] ksanjeevan/crnn-audio-classification: PyTorch Audio Classification: Urban Sounds

UrbanSound classification using Convolutional Recurrent Networks in PyTorch.



Generate text using OpenAI's new release of GPT-2 345M.

Papers & Publications

Few-Shot Unsupervised Image-to-Image Translation

Abstract: ….Drawing inspiration from the human capability of picking up the essence of a novel object from a small number of examples and generalizing from there, we seek a few-shot, unsupervised image-to-image translation algorithm that works on previously unseen target classes that are specified, at test time, only by a few example images. Our model achieves this few-shot generation capability by coupling an adversarial training scheme with a novel network design….

Visualizing the Consequences of Climate Change Using Cycle-Consistent Adversarial Networks

Abstract: We present a project that aims to generate images that depict accurate, vivid, and personalized outcomes of climate change using Cycle-Consistent Adversarial Networks (CycleGANs). By training our CycleGAN model on street-view images of houses before and after extreme weather events (e.g. floods, forest fires, etc.), we learn a mapping that can then be applied to images of locations that have not yet experienced these events. This visual transformation is paired with climate model predictions to assess likelihood and type of climate-related events in the long term (50 years) in order to bring the future closer in the viewers mind…