# Regressing Images

## Regressing 24 Hours in New Orleans

Regression is a widely applied technique in machine learning. An example of a regression problem would be the prediction of the length of a salmon as a function of its age and weight. Regression analysis is a statistical process for estimating the relationships among variables. Lets have some fun with it ;-)

# Experiment

A while back, **Andrej Karpathy** released **convnetjs**, a javascript based neural network library. It contains a demo which Andrej describes like this:

The demo treats the pixels of an image as a learning problem: it takes the (x,y) position on a grid and learns to predict the color at that point using regression to (r,g,b). It’s a bit like compression, since the image information is encoded in the weights of the network, but

almostcertainly not of practical kind :)

**This experiment test a regression based approach for video stylisation.** The following video was generated using **Stylize**** **by** ****Alec Radford****.** Alec extends Andrej’s implementation and uses a Random Forest Regressor. This experiment extends it to work on video. The source video is by **JacksGap****.**

## Regressing 24h in New Orleans

*If the youtube video is blocked, **try **vimeo*

## Selection of results

Heavy loading times, sorry!

## Selection of results: convnetJS

With bottleneck layer / dropout for extra fun.

# Final thoughts

Using machine learning techniques in unusual ways is fun, generating new image styles addictive. Realtime processing might be possible using GPUs (all CPU so far). A VR version could be fascinating ;-) All results in this experiment are reproducible with the linked open-source software. If setups are to cumbersome for you, try Mario Klingemann’s related “*Lowpoly Bot**”.*

Get in touch here: twitter.com/samim |http://samim.io

# Source Video

Special thanks to JackGap for the awesome source video! :-)