Hacker News new | past | comments | ask | show | jobs | submit login
Face detection for cats in JavaScript (harthur.github.com)
80 points by stevewilhelm on Sept 28, 2012 | hide | past | favorite | 36 comments



This work is based on a fairly old research paper. Since then, there has been significant progress on detecting dogs and cats more reliably.

Two prominent groups that are working on it are Andrew Zisserman's group at Oxford: www.robots.ox.ac.uk/~vgg/publications/2012/parkhi12a/parkhi12a.pdf

And Peter Belhumeur's group at Columbia University, who'll be presenting a new paper on detecting and recognizing dog breeds at the European Conference on Computer Vision (ECCV) in 2 weeks. [Peter was my PhD advisor]

More generally, many people in computer vision are getting excited about "fine-grained visual categorization," which is about classifying things at roughly the "species" level. This is in contrast to a lot of the previous computer vision literature, which either focused on generic categories (e.g., people vs animal vs car vs rocket-propelled-grenade) or specific object/instance recognition (e.g., face recognition).


Tom Cruise has definitely a catty smile - http://i.imgur.com/Mxbop.png



Failed to find cat on grass http://catoftheday.com/archive/2012/September/25.jpg

Failed to find cat yawning http://catoftheday.com/archive/2012/September/28.jpg

Found two cat faces with image of one houseplant and one cat http://catoftheday.com/archive/2012/September/28b.jpg

Success with Sphynx cat http://catoftheday.com/archive/2012/September/27.jpg


> Found two cat faces with image of one houseplant and one cat http://catoftheday.com/archive/2012/September/28b.jpg

Isn't there a second cat behind the houseplant?


I thought you were kidding at first. The algorithm is apparently better than humans at detecting cats.


This thing fails too often. I need accurate face detection for cats; and I need it now!


The drag-drop isn't working for me (Chrome Version 22.0.1229.79 on Unbuntu). It would be nice to have fallback to something simpler and more robust - i.e. a way to paste in a URL.


I can drag from nautilus to Chrome in kubuntu (yes, this is a weird combination), but not from Chrome to Chrome, for whatever reason.


Are you dragging in images from web pages? Or files from Nautilus?

Only the latter will work (speaking from experience)


Ah thanks, it's images from other web pages.

The rectangular border on the "drag an image here to find cats" area flashes like it's accepting the drop and then ... nothing.


Same with Chrome 22 on Windows 7.


Windows 7, chrome (newest), doesn't work, can't drag drop.

Uncaught TypeError: Cannot read property 'type' of undefined demo.js:64 handleFiles demo.js:64 (anonymous function) demo.js:48 f.event.dispatch jquery.js:3 h.handle.i


It looks like it loves ears on an upright cat. If the contrast between the angle of the ear and the background is clear it will eagerly draw boxes around them. I have lots of photos of my cat :


When you're bored, make a face detection app for cats.

Up next: Ass recognition for humans


This is not working for me, with the test images (or any other image). :\ Edit: Cannot read property 'type' of undefined. Demo.js:64


You have to click on the test images, rather than dragging the images onto the area!


Had the samme issue. Chrome 21 (and now 22 after checking version). Happened when I dragged and dropped an image from another Chrome window. Worked fine when dropping a local image.


Yeah, with local images it works fine.


What browser and platform are you using? I'm using Firefox and it's working fine for me.


Windows 8, chrome 22, opera next, opera 12.02, firefox 15. Hmm, actually in ie 10 it's doing it's stuff.


I have the same issue. I'm on Chrome + Ubuntu.



One of the test images with 4 cats on it was reported to have only 2 cats on it by the cat detector. Still, good work :)


It said some random part of my tshirt was a cat. Then I dragged a picture of a cat, and it said 0 cats. :(


Failed on a blatant cat face: http://d.pr/i/sAJK



I also fail to find a cat on this picture.


That's because Histogram of Oriented Gradients is not invariant to object orientation: http://en.wikipedia.org/wiki/Histogram_of_oriented_gradients... (second paragraph).

This example app was trained on upright cats.


Nope, still fails on a cat-upright version: http://i48.tinypic.com/2i0ejxl.jpg


That's not a cat, that's a dinosaur.


It's been completely ineffective so far, but it's a nice idea! :3


what algorithm do you use? Is there universal algorithm for face detect (or some kind of "object detect")?


For a good overview of methodology and terminology, start here http://www.idiap.ch/~rodrig/publications/pdf/rodrig-thesis-2...


Finally, the tool the internet has been screaming for since 1990!


Doesn't work.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: