Deep Learning Examples

Examples of Silverpond’s Deep Learning Prototypes

Fashion Space Explorer

Imagine if an AI could learn what fashion meant. You collaborate with it as an assistant fashion-designer to help you come up with new ideas for clothing structure and style. This dream is now a reality! The AI has been trained on Fashion-MNIST data to build up an 8-dimensional landscape of fasion that can be explored by users to find new garments and accessories. One you have found a suitable item, it can be colored with a second network in the style of a selection of artists.

The fashion-space explorer is a 2d embedding of the training data, and the points that people visited during Melbourne Knowledge Week 2018 (“MKW 2018”). You can read more about the MKW event here.

Dances Generated with Deep Learning

We built a model that takes in a given dance (perhaps captured via OpenPose), and responds with it’s own interpretation of that dance, done in it’s own style. These videos show the results of some of the generated dances from a meetup where we showcased the project. Sometimes the computer responds with an entire dance crew, where each individual dancer in the dance crew is responding to the same input dances, but dances it in their own unique style!

Model 1

Model 2

Virtual SCATS and counting traffic

Current traffic counting system such as SCATS can’t detect the types of vehicles in a road network. In this prototype we demonstrate the ability to count different types of vehicles using computer vision.

Object Tracking in Sports

We are using a CNN to identify a tennis ball and we implement a tracking to follow the ball.

Tracking human poses in our office

Silverpond experimenting with a CNN that detects human poses in a single web camera video feed. Possible use cases include pedestrian, limb and face detection.

Identifying gender and age

In this example, we demonstrate the ability to identify gender and age of a person. This system combines two models, one to detect the face and the other to calculate age and gender.

Designing Clothing and Accessories with Deep Learning

Using latent vectors to explore embedded fashion spaces, we demonstrate a multi-user input interface designed as a creative fashion design assistant:

Level 2
382 Little Collins Street
Melbourne VIC, 3000

Enter from McKillop Street

Contact us

(03) 9008 5922

Get the latest on machine learning and data science in your inbox each month