harthur / kittydar Goto Github PK

Face detection for cats in JavaScript - demo for TXJS 2012 talk

Home Page: http://harthur.github.com/kittydar

License: MIT License

JavaScript 98.62% Ruby 1.38%

kittydar's Introduction

Kittydar

Kittydar is short for kitty radar. Kittydar takes an image (canvas) and tells you the locations of all the cats in the image:

var cats = kittydar.detectCats(canvas);

console.log("there are", cats.length, "cats in this photo");

console.log(cats[0]);

// { x: 30, y: 200, width: 140, height: 140 }

Kittydar demo

Install

For node:

npm install kittydar

Or grab the browser file

Specifics

Kittydar takes a canvas element. In node you can get a Canvas object with node-canvas.

Kittydar will give an approximate rectangle around the cat's head. Each rectangle has an x and y for the top left corner, and a width and height of the rectangle.

How it works

Kittydar first chops the image up into many "windows" to test for the presence of a cat head. For each window, kittydar first extracts more tractable data from the image's data. Namely, it computes the Histogram of Orient Gradients descriptor of the image, using the hog-descriptor library. This data describes the directions of the edges in the image (where the image changes from light to dark and vice versa) and what strength they are. This data is a vector of numbers that is then fed into a neural network which gives a number from 0 to 1 on how likely the histogram data represents a cat.

The neural network (the JSON of which is located in this repo) has been pre-trained with thousands of photos of cat heads and their histograms, as well as thousands of non-cats. See the repo for the node training scripts.

Limitations

Kittydar will miss cats sometimes, and sometimes classify non-cats as cats. It's best at detecting upright cats that are facing forward, but it can handle a small tilt or turn in the head.

Kittydar isn't fast. It'll take a few seconds to find the cats in one image.

There's lots of room for improvement, so fork and send requests.

Propers

This informative reasearch paper: Cat Head Detection - How to Effectively Exploit Shape and Texture Features by Weiwei Zhang, Jian Sun, and Xiaoou Tang.
This off the hook dataset of cat images annotated with the locations of the cat's ears, eyes, and mouth.
@gdeglin for the name.

kittydar's People

Contributors

Stargazers

Watchers

Forkers

nivertech svankie mhejrati maxw3st jeresig cconstable pavlovcats lemonhall muthhus oahuja debus botriot gcirone jt5d amandarae neoscott jtmalinowski vj-chandera lujaw umaar codekk dblchu alessandraurso asauron projectskillz torreswil rommelosc mid9commander cat-alex trongnhantv mere lisendong bom-d-van andreydrv gyyinzjut dazkool pikachu123 iamironmansam dennyluan web5design tbranyen aficionado twocngdagz desertmonad alfsimen lgsunnyvale antonywu fedaykin m92o arnabkd zhuowater rockbot phpdotsql vackup shipoopi tsaizhenling gpiya grncdr niahm briefjudofox maserg jeffreycastellano ganhz nitrogen15 eiriklv juzefwt mvolz roborichard movb ryanliuntust keremtiryaki timoman alexandre-normand mirzap jwdai beeender mtco glompix akatashkov julien-moreau atinmathur sympatovit imbrianj nagyist prehensile fiestabonita gubagu blanksblanks ak4900 chapuzzo mrzeta tesslins ilkerc manelik girliemac tomjal richard512 robertjgabriel elmlea-young moshams

kittydar's Issues

node-canvas 1.0.1 fails to build (while the latest, v1.2.7 builds fine)

My node version is 0.12.6, running on Raspbian.

The version of node-canvas specified in the package.json fails to build, so I manually install the latest node-canvas and works just fine.
(But I need to make a work-around with the version of node-canvas. See Automattic/node-canvas#487 and
#20)

TypeError: Image or Canvas expected

Hi by executing:
node test.js
on a raspberry Pi with kittydar 1.6

where test.js contains the following JS:

var http = require('http'), fs = require('fs'),
Canvas = require('canvas');
kittyDar = require('kittydar');

fs.readFile('/var/www/kittenpush/web/images/userData/kittenimage.jpg', function(err, squid){
if (err) throw err;

img = new Canvas.Image;
img.src = squid;

var canvas = new Canvas(img.width, img.height);
var ctxxxx = canvas.getContext('2d');
ctxxxx.drawImage(img, 0, 0, img.width, img.height,0,0,img.width,img.height);

console.log(canvas);
var cats = kittyDar.detectCats(canvas);
console.log("there are", cats.length, "cats in this photo");
});

I'm ending up with an error insite the kittydar:

[Canvas 1600x1200]

/home/pi/node_modules/kittydar/kittydar.js:151
ctx.drawImage(canvas, 0, 0, canvas.width, canvas.height,
^
TypeError: Image or Canvas expected
at resizeCanvas (/home/pi/node_modules/kittydar/kittydar.js:151:7)
at Object.kittydar.scaleCanvas (/home/pi/node_modules/kittydar/kittydar.js:96:14)
at Object.kittydar.getAllSizes (/home/pi/node_modules/kittydar/kittydar.js:81:28)
at Object.kittydar.detectCats (/home/pi/node_modules/kittydar/kittydar.js:56:24)
at /home/pi/test.js:19:23
at fs.js:266:14
at Object.oncomplete (fs.js:107:15)

Im not sure maybe im just missing something on creating my canvas, but it seems like the kittydar package looses the canvas during execution?

Has anyone else this issue? Would be glad if someone could help.

Link broke in readme.md file

Hi - the link http://137.189.35.203/WebUI/CatDatabase/catData.html is dead.

Doesn't work for this image

Face recognition not working with Sphynx cats

I've discovered this software very seldom works with Sphynx cats (the only cat I would use this with). https://edgodanimal.files.wordpress.com/2012/05/0ad03169fae0a08c6ff7001f20c5fe46.jpg

it detects a puppy before a kitten

detected the dogs face rather than the kittens

new cat corpus for testing

https://github.com/maxogden/cats

hand photographed by me! all images contain cats and all images are the same dimensions and taken with the same camera

Implement coefficient adjustment / learning approach?

There is many false negatives and false positives with the demo.

We need something for easy tuning.

Two of the five cats looking forward not detected

The two right cats are looking straight forward and are not detected.

Also I found this similar issue, that should be moved to this repo. BrainJS/kittydar#1

uploaded an owl and it detected 1 cat :(

This is k-mart quality

This is not a cat

Original image: http://cl.ly/HP4G

Nevertheless, nice work

y u no find?

JS error when image is dropped onto the target

The error is:

ReferenceError: Can't find variable: FileReader

It's the same for any image.

Safari 5.1.7/Mac, no Adobe Flash on the system (if that's relevant)

can't detect cat on this image

can't detect cat on this image
http://sphotos-h.ak.fbcdn.net/hphotos-ak-ash3/550542_367679083310929_604435675_n.jpg

specific face recognition

Has anyone tried - what would be the route to take.

To identify a specific cat once the coordinates are returned?

Would be great for TNR purposes in tracking the individuals within a colony, and when a new cat has entered.

False positive. Pikachu and I palling around in Vegas.

First of all, very cool project!

Cartoon Pikachu not detected as cat. RL Pikachu = cat.

Original here: https://dl.dropbox.com/u/38059294/0115112240.jpg

How many cats?

Don't think it likes lots of cats facing forward
http://cl.ly/JmnM

Another detection issue

Interesting result for this image: http://twitpic.com/ayxztc

Does not detect cats with hats

I feel excluded

Cannot detect cat on this image

This is the cropped version of the positive image in the demo website. http://www1.picturepush.com/photo/a/11348659/220/11348659.jpg
It was recognized when I tested it with test-network.js but the JavaScript detector wasn't able to detect it. I'm really curious about the reason behind this.

1 cat detected

cat?

is there an implementation of ECCV08 paper?

nice work!
is this implementation follow the ECCV08 paper, Cat Head Detection - How to Effectively Exploit Shape and Texture Features by Weiwei Zhang, Jian Sun, and Xiaoou Tang? with fusion of texture and shape detectors?

I gothrough and find neural network things.

How long is train_svm supposed to take

Honestly, thank you for creating this project. It is more or less a learning experience for me. I read about neural network and svm in the past, but never fully grasp the ideas for me to successfully apply these concepts.

With the README, I followed the instructions. The train_network seems to finish the training within an hour or two, but I still need to play with it further (plus study the code).

As for train_svm, well, it has been running for almost 39 hours straight with no results

svm antony$ node train-svm.js -p ../POSITIVES -n ../NEGATIVES
training on 18867
feature size 3388

From my end, I can see node is using about 598Mb of memory with 101% of CPU (I guess it takes over one of the dual core). I can't tel if it is stuck in an infinite loop or it is actually doing something.

So has anyone tried train_svm? How long did it take for train-svm to finish?