kcimc

everything david (workergnome) said, and also: in the tile you point to, "swimming pool" is most interesting/recognizable feature, but most of the image is just "trees and asphalt". if you pick another image where the pool takes up more of the image, you'll find more of what you're looking for: http://nyc.terrapattern.com/?_ga=1.84865689.1830936426.14642...

I helped out with this project, happy to answer any questions. I was involved from the beginning, but my biggest contribution was on the deep learning side that does the tile-matching. I helped with the initial prototype using DIGITS, Caffe, and a bunch of Python. Then Aman Tiwari moved us to a more accurate 34-layer ResNet trained in TensorFlow, and a more efficient nearest neighbor lookup using a new implementation of CoverTree search.

as karpathy said, i really don't recommend it and i'm surprised it works at all. details here: https://gist.github.com/kylemcdonald/b02edbc33942a85856c8

The image is a composite of a photo by Tadao Cern. It was a joke, meant to look "too good to be true", but accepted without question: https://medium.com/@kcimc/finding-gogh-76ff90cbd408

In the artwork "The Secret Lives of Numbers"[1] by Golan Levin, et al. from 2002, scraping AltaVista with Perl for the numbers 0 to 1,000,000 returned results for every number (otherwise I'm sure this would be mentioned in the report[2]). So at least 7 figures have been omnipresent on the web for the last 12+ years.

[1] http://flong.com/projects/slon/ [2] http://flong.com/storage/pdf/reports/webnums_report.pdf

it was built with storm for processing all the listening data in realtime, node.js for relaying udp messages from storm to websockets, and d3.js for rendering the map along with the spotify web api for grabbing all the artist data, cover images, song previews.

usability isn't my strength, so this is all very helpful feedback. thanks!

creator here. sorry for the annoyance! original idea was to slowly fade the first song in, but i couldn't get the html5 audio api wrangled quickly enough for launch. the insanity of consecutive 3 second audio clips is definitely a part of the experience i didn't want anyone to miss, but sorry it couldn't have started more subtly.

thanks! all credit goes to the developers of d3.js though, i wouldn't have considered these projections if not for the example here http://bl.ocks.org/mbostock/3763057

the clips are spotify's preview mp3s, which are selected at random from the song. i know at least one very clever person at spotify who is trying to extract the 'hook' or 'recognisable part', and it worked really well last time i listened.