diff --git a/README.md b/README.md deleted file mode 100644 index 5b6c89d..0000000 --- a/README.md +++ /dev/null @@ -1,10 +0,0 @@ -Categorization of drawn cat and dog faces -========================================= - -Under [ml/](ml/) you can find my attempt to do it with machine learning, -where I decide myself how to extract which features and later use a simple -algorithm to classify. - -Under [tf/](tf/) you can find my attempt to do it with TensorFlow, which -is a neural network which extracts it's own features and neurons classify -the pictures. diff --git a/convert.sh b/convert.sh deleted file mode 100755 index 7e031d8..0000000 --- a/convert.sh +++ /dev/null @@ -1,10 +0,0 @@ -#!/bin/sh - -NEW="new_$1" -rm -rf $NEW -mkdir $NEW - -for i in `ls -1 $1` -do - convert $1/$i -adaptive-resize 200x200\> -size 200x200 xc:white +swap -gravity center -composite $NEW/$i -done diff --git a/images/bg_hr.png b/images/bg_hr.png new file mode 100644 index 0000000..7973bd6 Binary files /dev/null and b/images/bg_hr.png differ diff --git a/images/blacktocat.png b/images/blacktocat.png new file mode 100644 index 0000000..6e264fe Binary files /dev/null and b/images/blacktocat.png differ diff --git a/images/icon_download.png b/images/icon_download.png new file mode 100644 index 0000000..a2a287f Binary files /dev/null and b/images/icon_download.png differ diff --git a/images/sprite_download.png b/images/sprite_download.png new file mode 100644 index 0000000..f2babd5 Binary files /dev/null and b/images/sprite_download.png differ diff --git a/img/cat/1.png b/img/cat/1.png deleted file mode 100644 index f5d2fa2..0000000 Binary files a/img/cat/1.png and /dev/null differ diff --git a/img/cat/10.png b/img/cat/10.png deleted file mode 100644 index d5db6d1..0000000 Binary files a/img/cat/10.png and /dev/null differ diff --git a/img/cat/11.png b/img/cat/11.png deleted file mode 100644 index 7ec5c71..0000000 Binary files a/img/cat/11.png and /dev/null differ diff --git a/img/cat/12.png b/img/cat/12.png deleted file mode 100644 index 3d6faf4..0000000 Binary files a/img/cat/12.png and /dev/null differ diff --git a/img/cat/13.png b/img/cat/13.png deleted file mode 100644 index 97efb02..0000000 Binary files a/img/cat/13.png and /dev/null differ diff --git a/img/cat/14.png b/img/cat/14.png deleted file mode 100644 index 4fed457..0000000 Binary files a/img/cat/14.png and /dev/null differ diff --git a/img/cat/15.png b/img/cat/15.png deleted file mode 100644 index 3ab6f91..0000000 Binary files a/img/cat/15.png and /dev/null differ diff --git a/img/cat/16.png b/img/cat/16.png deleted file mode 100644 index 5ce067f..0000000 Binary files a/img/cat/16.png and /dev/null differ diff --git a/img/cat/17.png b/img/cat/17.png deleted file mode 100644 index ccca4a3..0000000 Binary files a/img/cat/17.png and /dev/null differ diff --git a/img/cat/18.png b/img/cat/18.png deleted file mode 100644 index d528bbe..0000000 Binary files a/img/cat/18.png and /dev/null differ diff --git a/img/cat/19.png b/img/cat/19.png deleted file mode 100644 index aff8023..0000000 Binary files a/img/cat/19.png and /dev/null differ diff --git a/img/cat/2.png b/img/cat/2.png deleted file mode 100644 index 81e417b..0000000 Binary files a/img/cat/2.png and /dev/null differ diff --git a/img/cat/20.png b/img/cat/20.png deleted file mode 100644 index fe2d584..0000000 Binary files a/img/cat/20.png and /dev/null differ diff --git a/img/cat/21.png b/img/cat/21.png deleted file mode 100644 index df90a5c..0000000 Binary files a/img/cat/21.png and /dev/null differ diff --git a/img/cat/22.png b/img/cat/22.png deleted file mode 100644 index 1443637..0000000 Binary files a/img/cat/22.png and /dev/null differ diff --git a/img/cat/23.png b/img/cat/23.png deleted file mode 100644 index 37aa2db..0000000 Binary files a/img/cat/23.png and /dev/null differ diff --git a/img/cat/24.png b/img/cat/24.png deleted file mode 100644 index ea1a480..0000000 Binary files a/img/cat/24.png and /dev/null differ diff --git a/img/cat/25.png b/img/cat/25.png deleted file mode 100644 index 50ce53f..0000000 Binary files a/img/cat/25.png and /dev/null differ diff --git a/img/cat/26.png b/img/cat/26.png deleted file mode 100644 index 6c5a51c..0000000 Binary files a/img/cat/26.png and /dev/null differ diff --git a/img/cat/27.png b/img/cat/27.png deleted file mode 100644 index 29c263b..0000000 Binary files a/img/cat/27.png and /dev/null differ diff --git a/img/cat/28.png b/img/cat/28.png deleted file mode 100644 index b5f0db8..0000000 Binary files a/img/cat/28.png and /dev/null differ diff --git a/img/cat/29.png b/img/cat/29.png deleted file mode 100644 index dd8f10c..0000000 Binary files a/img/cat/29.png and /dev/null differ diff --git a/img/cat/3.png b/img/cat/3.png deleted file mode 100644 index a3b1a53..0000000 Binary files a/img/cat/3.png and /dev/null differ diff --git a/img/cat/30.png b/img/cat/30.png deleted file mode 100644 index 843acb5..0000000 Binary files a/img/cat/30.png and /dev/null differ diff --git a/img/cat/31.png b/img/cat/31.png deleted file mode 100644 index 201a406..0000000 Binary files a/img/cat/31.png and /dev/null differ diff --git a/img/cat/32.png b/img/cat/32.png deleted file mode 100644 index 4c91e9c..0000000 Binary files a/img/cat/32.png and /dev/null differ diff --git a/img/cat/33.png b/img/cat/33.png deleted file mode 100644 index c61f711..0000000 Binary files a/img/cat/33.png and /dev/null differ diff --git a/img/cat/34.png b/img/cat/34.png deleted file mode 100644 index d5b689f..0000000 Binary files a/img/cat/34.png and /dev/null differ diff --git a/img/cat/35.png b/img/cat/35.png deleted file mode 100644 index 0f5d0de..0000000 Binary files a/img/cat/35.png and /dev/null differ diff --git a/img/cat/36.png b/img/cat/36.png deleted file mode 100644 index 2dccd69..0000000 Binary files a/img/cat/36.png and /dev/null differ diff --git a/img/cat/37.png b/img/cat/37.png deleted file mode 100644 index 8d03b22..0000000 Binary files a/img/cat/37.png and /dev/null differ diff --git a/img/cat/38.png b/img/cat/38.png deleted file mode 100644 index 612dac4..0000000 Binary files a/img/cat/38.png and /dev/null differ diff --git a/img/cat/39.png b/img/cat/39.png deleted file mode 100644 index 9d35a1e..0000000 Binary files a/img/cat/39.png and /dev/null differ diff --git a/img/cat/4.png b/img/cat/4.png deleted file mode 100644 index 4c086da..0000000 Binary files a/img/cat/4.png and /dev/null differ diff --git a/img/cat/40.png b/img/cat/40.png deleted file mode 100644 index a19812c..0000000 Binary files a/img/cat/40.png and /dev/null differ diff --git a/img/cat/41.png b/img/cat/41.png deleted file mode 100644 index 69a98b0..0000000 Binary files a/img/cat/41.png and /dev/null differ diff --git a/img/cat/42.png b/img/cat/42.png deleted file mode 100644 index df116d8..0000000 Binary files a/img/cat/42.png and /dev/null differ diff --git a/img/cat/43.png b/img/cat/43.png deleted file mode 100644 index bdd369c..0000000 Binary files a/img/cat/43.png and /dev/null differ diff --git a/img/cat/44.png b/img/cat/44.png deleted file mode 100644 index 1412a5a..0000000 Binary files a/img/cat/44.png and /dev/null differ diff --git a/img/cat/45.png b/img/cat/45.png deleted file mode 100644 index d66018c..0000000 Binary files a/img/cat/45.png and /dev/null differ diff --git a/img/cat/46.png b/img/cat/46.png deleted file mode 100644 index cccbdfe..0000000 Binary files a/img/cat/46.png and /dev/null differ diff --git a/img/cat/47.png b/img/cat/47.png deleted file mode 100644 index 54a6aed..0000000 Binary files a/img/cat/47.png and /dev/null differ diff --git a/img/cat/48.png b/img/cat/48.png deleted file mode 100644 index 42e850f..0000000 Binary files a/img/cat/48.png and /dev/null differ diff --git a/img/cat/49.png b/img/cat/49.png deleted file mode 100644 index f8cfda4..0000000 Binary files a/img/cat/49.png and /dev/null differ diff --git a/img/cat/5.png b/img/cat/5.png deleted file mode 100644 index e77bc3a..0000000 Binary files a/img/cat/5.png and /dev/null differ diff --git a/img/cat/50.png b/img/cat/50.png deleted file mode 100644 index 376b471..0000000 Binary files a/img/cat/50.png and /dev/null differ diff --git a/img/cat/51.png b/img/cat/51.png deleted file mode 100644 index 9b89c58..0000000 Binary files a/img/cat/51.png and /dev/null differ diff --git a/img/cat/52.png b/img/cat/52.png deleted file mode 100644 index e2f3a11..0000000 Binary files a/img/cat/52.png and /dev/null differ diff --git a/img/cat/54.png b/img/cat/54.png deleted file mode 100644 index e7ac183..0000000 Binary files a/img/cat/54.png and /dev/null differ diff --git a/img/cat/55.png b/img/cat/55.png deleted file mode 100644 index 27c3a6a..0000000 Binary files a/img/cat/55.png and /dev/null differ diff --git a/img/cat/56.png b/img/cat/56.png deleted file mode 100644 index f6ede4b..0000000 Binary files a/img/cat/56.png and /dev/null differ diff --git a/img/cat/57.png b/img/cat/57.png deleted file mode 100644 index 921f054..0000000 Binary files a/img/cat/57.png and /dev/null differ diff --git a/img/cat/58.png b/img/cat/58.png deleted file mode 100644 index 7a59712..0000000 Binary files a/img/cat/58.png and /dev/null differ diff --git a/img/cat/59.png b/img/cat/59.png deleted file mode 100644 index ec6ba49..0000000 Binary files a/img/cat/59.png and /dev/null differ diff --git a/img/cat/6.png b/img/cat/6.png deleted file mode 100644 index 7e0112e..0000000 Binary files a/img/cat/6.png and /dev/null differ diff --git a/img/cat/60.png b/img/cat/60.png deleted file mode 100644 index bf4a0c8..0000000 Binary files a/img/cat/60.png and /dev/null differ diff --git a/img/cat/61.png b/img/cat/61.png deleted file mode 100644 index 553a327..0000000 Binary files a/img/cat/61.png and /dev/null differ diff --git a/img/cat/62.png b/img/cat/62.png deleted file mode 100644 index 952732f..0000000 Binary files a/img/cat/62.png and /dev/null differ diff --git a/img/cat/63.png b/img/cat/63.png deleted file mode 100644 index 793a5f4..0000000 Binary files a/img/cat/63.png and /dev/null differ diff --git a/img/cat/64.png b/img/cat/64.png deleted file mode 100644 index b37c608..0000000 Binary files a/img/cat/64.png and /dev/null differ diff --git a/img/cat/65.png b/img/cat/65.png deleted file mode 100644 index 8410e61..0000000 Binary files a/img/cat/65.png and /dev/null differ diff --git a/img/cat/7.png b/img/cat/7.png deleted file mode 100644 index 1f7c5da..0000000 Binary files a/img/cat/7.png and /dev/null differ diff --git a/img/cat/8.png b/img/cat/8.png deleted file mode 100644 index f48b06d..0000000 Binary files a/img/cat/8.png and /dev/null differ diff --git a/img/cat/9.png b/img/cat/9.png deleted file mode 100644 index da7634b..0000000 Binary files a/img/cat/9.png and /dev/null differ diff --git a/img/dog/1.png b/img/dog/1.png deleted file mode 100644 index aea3814..0000000 Binary files a/img/dog/1.png and /dev/null differ diff --git a/img/dog/10.png b/img/dog/10.png deleted file mode 100644 index 8d79e2b..0000000 Binary files a/img/dog/10.png and /dev/null differ diff --git a/img/dog/11.png b/img/dog/11.png deleted file mode 100644 index 6f1e0e1..0000000 Binary files a/img/dog/11.png and /dev/null differ diff --git a/img/dog/12.png b/img/dog/12.png deleted file mode 100644 index aa4ef1d..0000000 Binary files a/img/dog/12.png and /dev/null differ diff --git a/img/dog/13.png b/img/dog/13.png deleted file mode 100644 index a6b84bb..0000000 Binary files a/img/dog/13.png and /dev/null differ diff --git a/img/dog/14.png b/img/dog/14.png deleted file mode 100644 index 81a7e2b..0000000 Binary files a/img/dog/14.png and /dev/null differ diff --git a/img/dog/15.png b/img/dog/15.png deleted file mode 100644 index 69aaca4..0000000 Binary files a/img/dog/15.png and /dev/null differ diff --git a/img/dog/16.png b/img/dog/16.png deleted file mode 100644 index ec4911e..0000000 Binary files a/img/dog/16.png and /dev/null differ diff --git a/img/dog/17.png b/img/dog/17.png deleted file mode 100644 index 50a5d18..0000000 Binary files a/img/dog/17.png and /dev/null differ diff --git a/img/dog/18.png b/img/dog/18.png deleted file mode 100644 index 4f4d834..0000000 Binary files a/img/dog/18.png and /dev/null differ diff --git a/img/dog/19.png b/img/dog/19.png deleted file mode 100644 index ebc4cb7..0000000 Binary files a/img/dog/19.png and /dev/null differ diff --git a/img/dog/2.png b/img/dog/2.png deleted file mode 100644 index ffa6fc0..0000000 Binary files a/img/dog/2.png and /dev/null differ diff --git a/img/dog/20.png b/img/dog/20.png deleted file mode 100644 index e94c52b..0000000 Binary files a/img/dog/20.png and /dev/null differ diff --git a/img/dog/21.png b/img/dog/21.png deleted file mode 100644 index b2eb6c3..0000000 Binary files a/img/dog/21.png and /dev/null differ diff --git a/img/dog/22.png b/img/dog/22.png deleted file mode 100644 index c55cc3f..0000000 Binary files a/img/dog/22.png and /dev/null differ diff --git a/img/dog/23.png b/img/dog/23.png deleted file mode 100644 index 4dd14c1..0000000 Binary files a/img/dog/23.png and /dev/null differ diff --git a/img/dog/24.png b/img/dog/24.png deleted file mode 100644 index 14c919a..0000000 Binary files a/img/dog/24.png and /dev/null differ diff --git a/img/dog/25.png b/img/dog/25.png deleted file mode 100644 index 2f4ee1b..0000000 Binary files a/img/dog/25.png and /dev/null differ diff --git a/img/dog/26.png b/img/dog/26.png deleted file mode 100644 index 0d70713..0000000 Binary files a/img/dog/26.png and /dev/null differ diff --git a/img/dog/27.png b/img/dog/27.png deleted file mode 100644 index cef11c5..0000000 Binary files a/img/dog/27.png and /dev/null differ diff --git a/img/dog/28.png b/img/dog/28.png deleted file mode 100644 index 6039765..0000000 Binary files a/img/dog/28.png and /dev/null differ diff --git a/img/dog/29.png b/img/dog/29.png deleted file mode 100644 index 55659a8..0000000 Binary files a/img/dog/29.png and /dev/null differ diff --git a/img/dog/3.png b/img/dog/3.png deleted file mode 100644 index ed9ba37..0000000 Binary files a/img/dog/3.png and /dev/null differ diff --git a/img/dog/30.png b/img/dog/30.png deleted file mode 100644 index 84f258a..0000000 Binary files a/img/dog/30.png and /dev/null differ diff --git a/img/dog/31.png b/img/dog/31.png deleted file mode 100644 index 7ad6c5d..0000000 Binary files a/img/dog/31.png and /dev/null differ diff --git a/img/dog/32.png b/img/dog/32.png deleted file mode 100644 index 00cf110..0000000 Binary files a/img/dog/32.png and /dev/null differ diff --git a/img/dog/33.png b/img/dog/33.png deleted file mode 100644 index 8b72422..0000000 Binary files a/img/dog/33.png and /dev/null differ diff --git a/img/dog/34.png b/img/dog/34.png deleted file mode 100644 index 4d15bd4..0000000 Binary files a/img/dog/34.png and /dev/null differ diff --git a/img/dog/35.png b/img/dog/35.png deleted file mode 100644 index e433d3e..0000000 Binary files a/img/dog/35.png and /dev/null differ diff --git a/img/dog/36.png b/img/dog/36.png deleted file mode 100644 index d8068d2..0000000 Binary files a/img/dog/36.png and /dev/null differ diff --git a/img/dog/37.png b/img/dog/37.png deleted file mode 100644 index db6b481..0000000 Binary files a/img/dog/37.png and /dev/null differ diff --git a/img/dog/38.png b/img/dog/38.png deleted file mode 100644 index db0eef2..0000000 Binary files a/img/dog/38.png and /dev/null differ diff --git a/img/dog/39.png b/img/dog/39.png deleted file mode 100644 index 4a0184e..0000000 Binary files a/img/dog/39.png and /dev/null differ diff --git a/img/dog/4.png b/img/dog/4.png deleted file mode 100644 index bd6de71..0000000 Binary files a/img/dog/4.png and /dev/null differ diff --git a/img/dog/40.png b/img/dog/40.png deleted file mode 100644 index 604d295..0000000 Binary files a/img/dog/40.png and /dev/null differ diff --git a/img/dog/41.png b/img/dog/41.png deleted file mode 100644 index a98a2ce..0000000 Binary files a/img/dog/41.png and /dev/null differ diff --git a/img/dog/43.png b/img/dog/43.png deleted file mode 100644 index 63cdd31..0000000 Binary files a/img/dog/43.png and /dev/null differ diff --git a/img/dog/44.png b/img/dog/44.png deleted file mode 100644 index 69ea750..0000000 Binary files a/img/dog/44.png and /dev/null differ diff --git a/img/dog/45.png b/img/dog/45.png deleted file mode 100644 index 29b40fa..0000000 Binary files a/img/dog/45.png and /dev/null differ diff --git a/img/dog/46.png b/img/dog/46.png deleted file mode 100644 index b58b32e..0000000 Binary files a/img/dog/46.png and /dev/null differ diff --git a/img/dog/47.png b/img/dog/47.png deleted file mode 100644 index 7684a17..0000000 Binary files a/img/dog/47.png and /dev/null differ diff --git a/img/dog/48.png b/img/dog/48.png deleted file mode 100644 index 75b8b6f..0000000 Binary files a/img/dog/48.png and /dev/null differ diff --git a/img/dog/49.png b/img/dog/49.png deleted file mode 100644 index fe738c9..0000000 Binary files a/img/dog/49.png and /dev/null differ diff --git a/img/dog/5.png b/img/dog/5.png deleted file mode 100644 index 9db7f19..0000000 Binary files a/img/dog/5.png and /dev/null differ diff --git a/img/dog/50.png b/img/dog/50.png deleted file mode 100644 index cbd10cd..0000000 Binary files a/img/dog/50.png and /dev/null differ diff --git a/img/dog/51.png b/img/dog/51.png deleted file mode 100644 index b823d74..0000000 Binary files a/img/dog/51.png and /dev/null differ diff --git a/img/dog/52.png b/img/dog/52.png deleted file mode 100644 index 1a00fad..0000000 Binary files a/img/dog/52.png and /dev/null differ diff --git a/img/dog/53.png b/img/dog/53.png deleted file mode 100644 index 7089f5d..0000000 Binary files a/img/dog/53.png and /dev/null differ diff --git a/img/dog/54.png b/img/dog/54.png deleted file mode 100644 index 4581dcc..0000000 Binary files a/img/dog/54.png and /dev/null differ diff --git a/img/dog/55.png b/img/dog/55.png deleted file mode 100644 index c9192dd..0000000 Binary files a/img/dog/55.png and /dev/null differ diff --git a/img/dog/56.png b/img/dog/56.png deleted file mode 100644 index 1b0267d..0000000 Binary files a/img/dog/56.png and /dev/null differ diff --git a/img/dog/57.png b/img/dog/57.png deleted file mode 100644 index 9a27022..0000000 Binary files a/img/dog/57.png and /dev/null differ diff --git a/img/dog/58.png b/img/dog/58.png deleted file mode 100644 index 1cfba41..0000000 Binary files a/img/dog/58.png and /dev/null differ diff --git a/img/dog/59.png b/img/dog/59.png deleted file mode 100644 index 0c40a55..0000000 Binary files a/img/dog/59.png and /dev/null differ diff --git a/img/dog/6.png b/img/dog/6.png deleted file mode 100644 index c3d0b81..0000000 Binary files a/img/dog/6.png and /dev/null differ diff --git a/img/dog/60.png b/img/dog/60.png deleted file mode 100644 index e069f57..0000000 Binary files a/img/dog/60.png and /dev/null differ diff --git a/img/dog/61.png b/img/dog/61.png deleted file mode 100644 index 4269f2d..0000000 Binary files a/img/dog/61.png and /dev/null differ diff --git a/img/dog/62.png b/img/dog/62.png deleted file mode 100644 index 57f858c..0000000 Binary files a/img/dog/62.png and /dev/null differ diff --git a/img/dog/63.png b/img/dog/63.png deleted file mode 100644 index e80c861..0000000 Binary files a/img/dog/63.png and /dev/null differ diff --git a/img/dog/64.png b/img/dog/64.png deleted file mode 100644 index 9ddf64a..0000000 Binary files a/img/dog/64.png and /dev/null differ diff --git a/img/dog/7.png b/img/dog/7.png deleted file mode 100644 index 6922a37..0000000 Binary files a/img/dog/7.png and /dev/null differ diff --git a/img/dog/8.png b/img/dog/8.png deleted file mode 100644 index b156b3a..0000000 Binary files a/img/dog/8.png and /dev/null differ diff --git a/img/dog/9.png b/img/dog/9.png deleted file mode 100644 index e241d0b..0000000 Binary files a/img/dog/9.png and /dev/null differ diff --git a/index.html b/index.html new file mode 100644 index 0000000..46c41c1 --- /dev/null +++ b/index.html @@ -0,0 +1,279 @@ + + + + + + + + + + + Cat vs. dog drawings categorization + + + + + +
+
+ View on GitHub + +

Cat vs. dog drawings categorization

+

A university project

+ +
+ Download this project as a .zip file + Download this project as a tar.gz file +
+
+
+ + +
+
+

+1. Introduction

+ +

+1.1. Goal

+ +

The goal of this project was to give a computer a drawing of either a cat's or a dog's face and let it recognize with high probability whether a cat or a dog is shown.

+ +

+1.2. Scope

+ +

First I thought that I would get lots of people to draw cat and dog faces for me, but I later realized that it was far too time consuming. Therefore I had to change the scope from recognizing random peoples drawings to recognizing my own drawings, which is obviously easier. Everything else did not change that much, I would just get better results.

+ +

+2. Preparation

+ +

+2.1. Drawing and taking a photo

+ +

The raw drawings

+ +

I drew eight A4 sheets of such cat and dog faces which resulted in 64 dog faces and 60 cat faces. Then I took pictures of them with my digital camera.

+ +

There was a huge difference in quality between the pictures I took with my iPhone 4 camera and the ones I took with my Nikon D5000. In fact I was not able to use the pictures I took with the iPhone because it was impossible to find straight lines in them.

+ +

You can see the result here, one with the iPhone image as a source and the other with the Nikon image:

+ +

iPhone vs. Nikon sample

+ +

+2.2. Photoshop

+ +

I cleaned up the drawings so it would be easier for the algorithm to find everything. I opened the pictures of the drawings in Photoshop and played with the contrast and brightness settings.

+ +

Then I cut out all the drawings from the big image and saved them as a black and white PNG images without dither.

+ +

Steps in Photoshop

+ +

+2.3. Resizing

+ +

I wrote a small shellscript which would take all pictures and resize them proportionally to a max width and height of 200 px. It also fills up the missing borders with a white background color. To do that I used the ImageMagick software suite:

+ +
#!/bin/sh
+
+NEW="new_$1"
+rm -rf $NEW
+mkdir $NEW
+
+for i in `ls -1 $1`
+do
+    convert $1/$i \
+        -adaptive-resize 200x200\> \
+        -size 200x200 xc:white +swap \
+        -gravity center \
+        -composite \
+        $NEW/$i
+done
+ +

After that all the images had uniform sizes and colors so that I was able to compare them in a meaningful way.

+ +

+3. Feature extraction

+ +

The next step was to extract the features from the images. In other words find things in the pictures that would be unique enough to make a difference between cats and dogs but broad enough so that all dogs would fall into one category and all cats into the other.

+ +

+3.1. Straight lines

+ +

The first thing which came to mind was counting and doing other stuff with straight lines in the image.

+ +

+3.1.1 Canny edge detector

+ +

I used an edge detector algorithm called Canny to preprocess the images which - as the name implies - finds edges in images. Because of my preparation with Photoshop it was quite easy for it to find them. It is not easy to see this step with my drawings, so here is a picture of how it looks like when you do this with a photo instead:

+ +

Canny on a photo from Wikipedia

+ +

It basically removes noise with a gausian filter and then finds the intentisty gradians of the image with help of some trigonometry.

+ +

I did not implement the algorithm myself, instead I used the often used OpenCV implementation.

+ +

+3.1.2 Hough transform

+ +

To find the lines I used the Hough transform algorithm. The red lines are those which the Hough transform algorithm found in the example picture:

+ +

Hough lines

+ +

It essentially groups edges, which can be imperfect, to object candidates by performing an explicit voting procedure. Detecting straight lines can be done by describing them as y = mx + b where m is the slope of the line and b is the intercept. The line is not represented by descrete points (x1,y1)(x2,y2) but instead as a point(x,y) in the parameter space, which makes detection of lines, which are a bit off, possible. In practice it is still more complicated, please read the Wikipedia article about it.

+ +

I did not implement it myself but used the often used and tested probabilistic OpenCV implementation.

+ +

+3.2. Line features

+ +

I extracted these features from the lines:

+ +

+3.3. Other features

+ +

I also extracted the amount of black pixels in the image to use it as a possible feature which was not using the extracted lines.

+ +

+4. k-nearest neighbor algorithm

+ +

I chose to use the k-Nearest Neighbors algorithm which only locally looks at the neighbors of the document in a radius predefined by the user. It assumes that the document is of the same category as the highest number of neighbors within this radius. +In the following figure you can see that depending if the user choses k = 3, as shown by the solid line, the algorithm will conclude that the document in the center (green smiley) is of the type triangle because most of this three neighbors are triangles. If on the other hand the user choses k = 7, as shown by the dotted line, then the amount of neighbors which are rectangles is greater as the amount of neighbors which are triangles, so it concludes that the smiley is of type rectangle.

+ +

k-Nearest Neighbours as a graphic

+ +

In the picture above you see how it would look with two dimensions. I have been using four features so the algorithm had to check the distance to the neighbours in four dimensions. This is not really more difficult, it is just more to calculate.

+ +

+5. Results

+ +

The results were quite encouraging, I assume it is because I only used one style to draw the dogs and one style to draw the cats.

+ +

+5.1. k-fold Cross-validation

+ +

I used 10 fold cross-validation for every test I did, which means that I used 90% of the available data for the learning algorithms and then the remaining 10% to test how they performed. I repeated this ten times until all data had been used for testing once.

+ +

+5.2. Results with all features

+ +

When I used all of the features and three nearest neighbours I got amazing 100% accuracy, which was kind of suspect because that normally means that you most probably did something wrong.

+ +

+5.3. Results with a reduced feature set

+ +

Therefore I tried to reduce the features to check if it would perform worse.

+ +
    +
  1. When I removed the information about the amount of black pixels basically nothing happened.
    2. +
  2. When I removed the information about the amount of lines and average length at least I got a couple of wrong categorized images, the accuracy went down to 95%.
    3. +
  3. When I removed the information about the average angle of the lines, that was when I got significant errors. The accuracy dropped down to about 60%, which is still better then pure chance.
    4. +

So it seems like the best feature to detect cat and dog face drawings done by me was the average angle of the straight lines in the image.

+ +

+6. Future study

+ +

The most important next step would be to gather many more drawings done by other people who use other styles to draw cat and dog faces.

+ +

Then it would be interesting to use other learning algorithms like Bayes, Perceptron, etc.

+ +

And then it would be interesting to use this approach on photos of real cats and dogs.

+ +

+7. Code

+ +
#!/usr/bin/env python
+
+import cv2, cv, sys, math, os, numpy
+from scipy.spatial import KDTree
+
+def extractFeatures(label):
+
+    directory = "img/" + label + "/"
+
+    features = []
+
+    for fn in os.listdir(directory):
+
+        img = cv2.imread(directory + fn, 0)
+
+        # find edges
+        canny = cv2.Canny(img, 50, 100)
+
+        # find colored
+        black_pixels = numpy.count_nonzero(img)
+
+        # find lines lines
+        lines = cv2.HoughLinesP(canny, 1, math.pi/360, 5, None, 10, 1)
+
+        lengths = []
+        angles = []
+        try:
+            for line in lines[0]:
+                x1, y1, x2, y2 = line
+
+                # Pythagoras
+                a2 = math.pow((x1-x2), 2)
+                b2 = math.pow((y1-y2), 2)
+                length = int(math.sqrt(a2 + b2))
+                lengths.append(length)
+
+                angle = int(math.degrees(math.atan((y1-y2) / (x1-x2))))
+                angles.append(angle)
+        except:
+            pass
+
+        # print out everything
+        lines_count = len(lengths)
+        mid_length = sum(lengths) / lines_count
+        mid_angle = sum(angles) / lines_count
+
+        features.append([
+            [lines_count, mid_length, mid_angle, black_pixels],
+            label
+        ])
+
+    return features
+
+
+if __name__ == "__main__":
+    cats = extractFeatures("cat")
+    dogs = extractFeatures("dog")
+
+    test_count = 5
+
+    test_data = dogs[:test_count] + cats[:test_count] 
+    test_labels = map(lambda a: a[1], test_data)
+    test_features = map(lambda a: a[0], test_data)
+
+    data = cats[test_count:] + dogs[test_count:]
+    labels = map(lambda a: a[1], data)
+    features = map(lambda a: a[0], data)
+
+    tree = KDTree(features)
+
+    for t in xrange(0, test_count * 2):
+        d, i = tree.query(test_features[t], k=3)
+        print "-"
+        for j in xrange(0, len(i)):
+            print test_labels[t] + " is a " + labels[i[j]]
+
+
+
+ + + + + + + + diff --git a/javascripts/main.js b/javascripts/main.js new file mode 100644 index 0000000..d8135d3 --- /dev/null +++ b/javascripts/main.js @@ -0,0 +1 @@ +console.log('This would be the main JS file.'); diff --git a/ml/Presentation-Swedish.odp b/ml/Presentation-Swedish.odp deleted file mode 100644 index b3f14a1..0000000 Binary files a/ml/Presentation-Swedish.odp and /dev/null differ diff --git a/ml/README.md b/ml/README.md deleted file mode 100644 index ba8af1c..0000000 --- a/ml/README.md +++ /dev/null @@ -1,210 +0,0 @@ -# 1. Introduction - -## 1.1. Goal - -The goal of this project was to give a computer a drawing of either a cat's or a dog's face and let it recognize with high probability whether a cat or a dog is shown. - -## 1.2. Scope - -First I thought that I would get lots of people to draw cat and dog faces for me, but I later realized that it was far too time consuming. Therefore I had to change the scope from recognizing random peoples drawings to recognizing my own drawings, which is obviously easier. Everything else did not change that much, I would just get better results. - -# 2. Preparation - -## 2.1. Drawing and taking a photo - -![The raw drawings](https://jeena.net/images/2013/catdog/drawing-taking-photo.jpg) - -I drew eight A4 sheets of such cat and dog faces which resulted in 64 dog faces and 60 cat faces. Then I took pictures of them with my digital camera. - -There was a huge difference in quality between the pictures I took with my iPhone 4 camera and the ones I took with my Nikon D5000. In fact I was not able to use the pictures I took with the iPhone because it was impossible to find straight lines in them. - -You can see the result here, one with the iPhone image as a source and the other with the Nikon image: - -![iPhone vs. Nikon sample](https://jeena.net/images/2013/catdog/iphone-sample.jpg) - -## 2.2. Photoshop - -I cleaned up the drawings so it would be easier for the algorithm to find everything. I opened the pictures of the drawings in Photoshop and played with the contrast and brightness settings. - -Then I cut out all the drawings from the big image and saved them as a black and white PNG images without dither. - -![Steps in Photoshop](https://jeena.net/images/2013/catdog/photoshop.jpg) - -## 2.3. Resizing - -I wrote a small shellscript which would take all pictures and resize them proportionally to a max width and height of 200 px. It also fills up the missing borders with a white background color. To do that I used the [ImageMagick](www.imagemagick.org) software suite: - -
#!/bin/sh
-
-NEW="new_$1"
-rm -rf $NEW
-mkdir $NEW
-
-for i in `ls -1 $1`
-do
-	convert $1/$i \
-		-adaptive-resize 200x200\> \
-		-size 200x200 xc:white +swap \
-		-gravity center \
-		-composite \
-		$NEW/$i
-done
- -After that all the images had uniform sizes and colors so that I was able to compare them in a meaningful way. - -# 3. Feature extraction - -The next step was to extract the features from the images. In other words find things in the pictures that would be unique enough to make a difference between cats and dogs but broad enough so that all dogs would fall into one category and all cats into the other. - -## 3.1. Straight lines - -The first thing which came to mind was counting and doing other stuff with straight lines in the image. - -### 3.1.1 Canny edge detector - -I used an edge detector algorithm called Canny to preprocess the images which - as the name implies - finds edges in images. Because of my preparation with Photoshop it was quite easy for it to find them. It is not easy to see this step with my drawings, so here is a picture of how it looks like when you do this with a photo instead: - -![Canny on a photo from Wikipedia](https://jeena.net/images/2013/catdog/canny.jpg) - -It basically removes noise with a gausian filter and then finds the intentisty gradians of the image with help of some trigonometry. - -I did not implement the algorithm myself, instead I used the often used [OpenCV implementation](http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/canny_detector/canny_detector.html). - -### 3.1.2 Hough transform - -To find the lines I used the [Hough transform](https://en.wikipedia.org/wiki/Hough_transform) algorithm. The red lines are those which the Hough transform algorithm found in the example picture: - -![Hough lines](https://jeena.net/images/2013/catdog/hough.png) - -It essentially groups edges, which can be imperfect, to object candidates by performing an explicit voting procedure. Detecting straight lines can be done by describing them as y = mx + b where m is the slope of the line and b is the intercept. The line is not represented by descrete points (x1,y1)(x2,y2) but instead as a point(x,y) in the parameter space, which makes detection of lines, which are a bit off, possible. In practice it is still more complicated, please read the [Wikipedia article](https://en.wikipedia.org/wiki/Hough_transform) about it. - -I did not implement it myself but used the often used and tested probabilistic [OpenCV implementation](http://docs.opencv.org/modules/imgproc/doc/feature_detection.html?highlight=houghlinesp#houghlinesp). - -## 3.2. Line features - -I extracted these features from the lines: - -- amount of lines -- average length of lines -- average angle of lines - -## 3.3. Other features - -I also extracted the amount of black pixels in the image to use it as a possible feature which was not using the extracted lines. - -# 4. _k_-nearest neighbor algorithm - -I chose to use the _k_-Nearest Neighbors algorithm which only locally looks at the neighbors of the document in a radius predefined by the user. It assumes that the document is of the same category as the highest number of neighbors within this radius. -In the following figure you can see that depending if the user choses k = 3, as shown by the solid line, the algorithm will conclude that the document in the center (green smiley) is of the type triangle because most of this three neighbors are triangles. If on the other hand the user choses k = 7, as shown by the dotted line, then the amount of neighbors which are rectangles is greater as the amount of neighbors which are triangles, so it concludes that the smiley is of type rectangle. - -![k-Nearest Neighbours as a graphic](https://jeena.net/images/2013/catdog/k-nearest-neighbours.png) - -In the picture above you see how it would look with two dimensions. I have been using four features so the algorithm had to check the distance to the neighbours in four dimensions. This is not really more difficult, it is just more to calculate. - -# 5. Results - -The results were quite encouraging, I assume it is because I only used one style to draw the dogs and one style to draw the cats. - -## 5.1. k-fold Cross-validation - -I used 10 fold cross-validation for every test I did, which means that I used 90% of the available data for the learning algorithms and then the remaining 10% to test how they performed. I repeated this ten times until all data had been used for testing once. - -## 5.2. Results with all features - -When I used all of the features and three nearest neighbours I got amazing 100% accuracy, which was kind of suspect because that normally means that you most probably did something wrong. - -## 5.3. Results with a reduced feature set - -Therefore I tried to reduce the features to check if it would perform worse. - -1. When I removed the information about the amount of black pixels basically nothing happened. -2. When I removed the information about the amount of lines and average length at least I got a couple of wrong categorized images, the accuracy went down to 95%. -3. When I removed the information about the average angle of the lines, that was when I got significant errors. The accuracy dropped down to about 60%, which is still better then pure chance. - -So it seems like the best feature to detect cat and dog face drawings done by me was the average angle of the straight lines in the image. - -# 6. Future study - -The most important next step would be to gather many more drawings done by other people who use other styles to draw cat and dog faces. - -Then it would be interesting to use other learning algorithms like Bayes, Perceptron, etc. - -And then it would be interesting to use this approach on photos of real cats and dogs. - -# 7. Code - - #!/usr/bin/env python - - import cv2, cv, sys, math, os, numpy - from scipy.spatial import KDTree - - def extractFeatures(label): - - directory = "img/" + label + "/" - - features = [] - - for fn in os.listdir(directory): - - img = cv2.imread(directory + fn, 0) - - # find edges - canny = cv2.Canny(img, 50, 100) - - # find colored - black_pixels = numpy.count_nonzero(img) - - # find lines lines - lines = cv2.HoughLinesP(canny, 1, math.pi/360, 5, None, 10, 1) - - lengths = [] - angles = [] - try: - for line in lines[0]: - x1, y1, x2, y2 = line - - # Pythagoras - a2 = math.pow((x1-x2), 2) - b2 = math.pow((y1-y2), 2) - length = int(math.sqrt(a2 + b2)) - lengths.append(length) - - angle = int(math.degrees(math.atan((y1-y2) / (x1-x2)))) - angles.append(angle) - except: - pass - - # print out everything - lines_count = len(lengths) - mid_length = sum(lengths) / lines_count - mid_angle = sum(angles) / lines_count - - features.append([ - [lines_count, mid_length, mid_angle, black_pixels], - label - ]) - - return features - - - if __name__ == "__main__": - cats = extractFeatures("cat") - dogs = extractFeatures("dog") - - test_count = 5 - - test_data = dogs[:test_count] + cats[:test_count] - test_labels = map(lambda a: a[1], test_data) - test_features = map(lambda a: a[0], test_data) - - data = cats[test_count:] + dogs[test_count:] - labels = map(lambda a: a[1], data) - features = map(lambda a: a[0], data) - - tree = KDTree(features) - - for t in xrange(0, test_count * 2): - d, i = tree.query(test_features[t], k=3) - print "-" - for j in xrange(0, len(i)): - print test_labels[t] + " is a " + labels[i[j]] diff --git a/ml/classify.py b/ml/classify.py deleted file mode 100755 index 3168d70..0000000 --- a/ml/classify.py +++ /dev/null @@ -1,76 +0,0 @@ -#!/usr/bin/env python - -import cv2, cv, sys, math, os, numpy -from scipy.spatial import KDTree - -def extractFeatures(label): - - directory = "img/" + label + "/" - - features = [] - - for fn in os.listdir(directory): - - img = cv2.imread(directory + fn, 0) - - # find edges - canny = cv2.Canny(img, 50, 100) - - # find colored - black_pixels = numpy.count_nonzero(img) - - # find lines lines - lines = cv2.HoughLinesP(canny, 1, math.pi/360, 5, None, 10, 1) - - lengths = [] - angles = [] - try: - for line in lines[0]: - x1, y1, x2, y2 = line - - # Pythagoras - a2 = math.pow((x1-x2), 2) - b2 = math.pow((y1-y2), 2) - length = int(math.sqrt(a2 + b2)) - lengths.append(length) - - angle = int(math.degrees(math.atan((y1-y2) / (x1-x2)))) - angles.append(angle) - except: - pass - - # print out everything - lines_count = len(lengths) - mid_length = sum(lengths) / lines_count - mid_angle = sum(angles) / lines_count - - features.append([ - [lines_count, mid_length, mid_angle, black_pixels], - label - ]) - - return features - - -if __name__ == "__main__": - cats = extractFeatures("cat") - dogs = extractFeatures("dog") - - test_count = 5 - - test_data = dogs[:test_count] + cats[:test_count] - test_labels = map(lambda a: a[1], test_data) - test_features = map(lambda a: a[0], test_data) - - data = cats[test_count:] + dogs[test_count:] - labels = map(lambda a: a[1], data) - features = map(lambda a: a[0], data) - - tree = KDTree(features) - - for t in xrange(0, test_count * 2): - d, i = tree.query(test_features[t], k=3) - print "-" - for j in xrange(0, len(i)): - print test_labels[t] + " is predicted to be a " + labels[i[j]] - diff --git a/ml/report.md b/ml/report.md deleted file mode 100644 index ba8af1c..0000000 --- a/ml/report.md +++ /dev/null @@ -1,210 +0,0 @@ -# 1. Introduction - -## 1.1. Goal - -The goal of this project was to give a computer a drawing of either a cat's or a dog's face and let it recognize with high probability whether a cat or a dog is shown. - -## 1.2. Scope - -First I thought that I would get lots of people to draw cat and dog faces for me, but I later realized that it was far too time consuming. Therefore I had to change the scope from recognizing random peoples drawings to recognizing my own drawings, which is obviously easier. Everything else did not change that much, I would just get better results. - -# 2. Preparation - -## 2.1. Drawing and taking a photo - -![The raw drawings](https://jeena.net/images/2013/catdog/drawing-taking-photo.jpg) - -I drew eight A4 sheets of such cat and dog faces which resulted in 64 dog faces and 60 cat faces. Then I took pictures of them with my digital camera. - -There was a huge difference in quality between the pictures I took with my iPhone 4 camera and the ones I took with my Nikon D5000. In fact I was not able to use the pictures I took with the iPhone because it was impossible to find straight lines in them. - -You can see the result here, one with the iPhone image as a source and the other with the Nikon image: - -![iPhone vs. Nikon sample](https://jeena.net/images/2013/catdog/iphone-sample.jpg) - -## 2.2. Photoshop - -I cleaned up the drawings so it would be easier for the algorithm to find everything. I opened the pictures of the drawings in Photoshop and played with the contrast and brightness settings. - -Then I cut out all the drawings from the big image and saved them as a black and white PNG images without dither. - -![Steps in Photoshop](https://jeena.net/images/2013/catdog/photoshop.jpg) - -## 2.3. Resizing - -I wrote a small shellscript which would take all pictures and resize them proportionally to a max width and height of 200 px. It also fills up the missing borders with a white background color. To do that I used the [ImageMagick](www.imagemagick.org) software suite: - -
#!/bin/sh
-
-NEW="new_$1"
-rm -rf $NEW
-mkdir $NEW
-
-for i in `ls -1 $1`
-do
-	convert $1/$i \
-		-adaptive-resize 200x200\> \
-		-size 200x200 xc:white +swap \
-		-gravity center \
-		-composite \
-		$NEW/$i
-done
- -After that all the images had uniform sizes and colors so that I was able to compare them in a meaningful way. - -# 3. Feature extraction - -The next step was to extract the features from the images. In other words find things in the pictures that would be unique enough to make a difference between cats and dogs but broad enough so that all dogs would fall into one category and all cats into the other. - -## 3.1. Straight lines - -The first thing which came to mind was counting and doing other stuff with straight lines in the image. - -### 3.1.1 Canny edge detector - -I used an edge detector algorithm called Canny to preprocess the images which - as the name implies - finds edges in images. Because of my preparation with Photoshop it was quite easy for it to find them. It is not easy to see this step with my drawings, so here is a picture of how it looks like when you do this with a photo instead: - -![Canny on a photo from Wikipedia](https://jeena.net/images/2013/catdog/canny.jpg) - -It basically removes noise with a gausian filter and then finds the intentisty gradians of the image with help of some trigonometry. - -I did not implement the algorithm myself, instead I used the often used [OpenCV implementation](http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/canny_detector/canny_detector.html). - -### 3.1.2 Hough transform - -To find the lines I used the [Hough transform](https://en.wikipedia.org/wiki/Hough_transform) algorithm. The red lines are those which the Hough transform algorithm found in the example picture: - -![Hough lines](https://jeena.net/images/2013/catdog/hough.png) - -It essentially groups edges, which can be imperfect, to object candidates by performing an explicit voting procedure. Detecting straight lines can be done by describing them as y = mx + b where m is the slope of the line and b is the intercept. The line is not represented by descrete points (x1,y1)(x2,y2) but instead as a point(x,y) in the parameter space, which makes detection of lines, which are a bit off, possible. In practice it is still more complicated, please read the [Wikipedia article](https://en.wikipedia.org/wiki/Hough_transform) about it. - -I did not implement it myself but used the often used and tested probabilistic [OpenCV implementation](http://docs.opencv.org/modules/imgproc/doc/feature_detection.html?highlight=houghlinesp#houghlinesp). - -## 3.2. Line features - -I extracted these features from the lines: - -- amount of lines -- average length of lines -- average angle of lines - -## 3.3. Other features - -I also extracted the amount of black pixels in the image to use it as a possible feature which was not using the extracted lines. - -# 4. _k_-nearest neighbor algorithm - -I chose to use the _k_-Nearest Neighbors algorithm which only locally looks at the neighbors of the document in a radius predefined by the user. It assumes that the document is of the same category as the highest number of neighbors within this radius. -In the following figure you can see that depending if the user choses k = 3, as shown by the solid line, the algorithm will conclude that the document in the center (green smiley) is of the type triangle because most of this three neighbors are triangles. If on the other hand the user choses k = 7, as shown by the dotted line, then the amount of neighbors which are rectangles is greater as the amount of neighbors which are triangles, so it concludes that the smiley is of type rectangle. - -![k-Nearest Neighbours as a graphic](https://jeena.net/images/2013/catdog/k-nearest-neighbours.png) - -In the picture above you see how it would look with two dimensions. I have been using four features so the algorithm had to check the distance to the neighbours in four dimensions. This is not really more difficult, it is just more to calculate. - -# 5. Results - -The results were quite encouraging, I assume it is because I only used one style to draw the dogs and one style to draw the cats. - -## 5.1. k-fold Cross-validation - -I used 10 fold cross-validation for every test I did, which means that I used 90% of the available data for the learning algorithms and then the remaining 10% to test how they performed. I repeated this ten times until all data had been used for testing once. - -## 5.2. Results with all features - -When I used all of the features and three nearest neighbours I got amazing 100% accuracy, which was kind of suspect because that normally means that you most probably did something wrong. - -## 5.3. Results with a reduced feature set - -Therefore I tried to reduce the features to check if it would perform worse. - -1. When I removed the information about the amount of black pixels basically nothing happened. -2. When I removed the information about the amount of lines and average length at least I got a couple of wrong categorized images, the accuracy went down to 95%. -3. When I removed the information about the average angle of the lines, that was when I got significant errors. The accuracy dropped down to about 60%, which is still better then pure chance. - -So it seems like the best feature to detect cat and dog face drawings done by me was the average angle of the straight lines in the image. - -# 6. Future study - -The most important next step would be to gather many more drawings done by other people who use other styles to draw cat and dog faces. - -Then it would be interesting to use other learning algorithms like Bayes, Perceptron, etc. - -And then it would be interesting to use this approach on photos of real cats and dogs. - -# 7. Code - - #!/usr/bin/env python - - import cv2, cv, sys, math, os, numpy - from scipy.spatial import KDTree - - def extractFeatures(label): - - directory = "img/" + label + "/" - - features = [] - - for fn in os.listdir(directory): - - img = cv2.imread(directory + fn, 0) - - # find edges - canny = cv2.Canny(img, 50, 100) - - # find colored - black_pixels = numpy.count_nonzero(img) - - # find lines lines - lines = cv2.HoughLinesP(canny, 1, math.pi/360, 5, None, 10, 1) - - lengths = [] - angles = [] - try: - for line in lines[0]: - x1, y1, x2, y2 = line - - # Pythagoras - a2 = math.pow((x1-x2), 2) - b2 = math.pow((y1-y2), 2) - length = int(math.sqrt(a2 + b2)) - lengths.append(length) - - angle = int(math.degrees(math.atan((y1-y2) / (x1-x2)))) - angles.append(angle) - except: - pass - - # print out everything - lines_count = len(lengths) - mid_length = sum(lengths) / lines_count - mid_angle = sum(angles) / lines_count - - features.append([ - [lines_count, mid_length, mid_angle, black_pixels], - label - ]) - - return features - - - if __name__ == "__main__": - cats = extractFeatures("cat") - dogs = extractFeatures("dog") - - test_count = 5 - - test_data = dogs[:test_count] + cats[:test_count] - test_labels = map(lambda a: a[1], test_data) - test_features = map(lambda a: a[0], test_data) - - data = cats[test_count:] + dogs[test_count:] - labels = map(lambda a: a[1], data) - features = map(lambda a: a[0], data) - - tree = KDTree(features) - - for t in xrange(0, test_count * 2): - d, i = tree.query(test_features[t], k=3) - print "-" - for j in xrange(0, len(i)): - print test_labels[t] + " is a " + labels[i[j]] diff --git a/params.json b/params.json new file mode 100644 index 0000000..96c81bc --- /dev/null +++ b/params.json @@ -0,0 +1 @@ +{"name":"Cat vs. dog drawings categorization","tagline":"A university project","body":"# 1. Introduction\r\n\r\n## 1.1. Goal\r\n\r\nThe goal of this project was to give a computer a drawing of either a cat's or a dog's face and let it recognize with high probability whether a cat or a dog is shown.\r\n\r\n## 1.2. Scope\r\n\r\nFirst I thought that I would get lots of people to draw cat and dog faces for me, but I later realized that it was far too time consuming. Therefore I had to change the scope from recognizing random peoples drawings to recognizing my own drawings, which is obviously easier. Everything else did not change that much, I would just get better results.\r\n\r\n# 2. Preparation\r\n\r\n## 2.1. Drawing and taking a photo\r\n\r\n![The raw drawings](https://jeena.net/images/2013/catdog/drawing-taking-photo.jpg)\r\n\r\nI drew eight A4 sheets of such cat and dog faces which resulted in 64 dog faces and 60 cat faces. Then I took pictures of them with my digital camera.\r\n\r\nThere was a huge difference in quality between the pictures I took with my iPhone 4 camera and the ones I took with my Nikon D5000. In fact I was not able to use the pictures I took with the iPhone because it was impossible to find straight lines in them.\r\n\r\nYou can see the result here, one with the iPhone image as a source and the other with the Nikon image:\r\n\r\n![iPhone vs. Nikon sample](https://jeena.net/images/2013/catdog/iphone-sample.jpg)\r\n\r\n## 2.2. Photoshop\r\n\r\nI cleaned up the drawings so it would be easier for the algorithm to find everything. I opened the pictures of the drawings in Photoshop and played with the contrast and brightness settings.\r\n\r\nThen I cut out all the drawings from the big image and saved them as a black and white PNG images without dither.\r\n\r\n![Steps in Photoshop](https://jeena.net/images/2013/catdog/photoshop.jpg)\r\n\r\n## 2.3. Resizing\r\n\r\nI wrote a small shellscript which would take all pictures and resize them proportionally to a max width and height of 200 px. It also fills up the missing borders with a white background color. To do that I used the [ImageMagick](www.imagemagick.org) software suite:\r\n\r\n
#!/bin/sh\r\n\r\nNEW=\"new_$1\"\r\nrm -rf $NEW\r\nmkdir $NEW\r\n\r\nfor i in `ls -1 $1`\r\ndo\r\n\tconvert $1/$i \\\r\n\t\t-adaptive-resize 200x200\\> \\\r\n\t\t-size 200x200 xc:white +swap \\\r\n\t\t-gravity center \\\r\n\t\t-composite \\\r\n\t\t$NEW/$i\r\ndone
\r\n\r\nAfter that all the images had uniform sizes and colors so that I was able to compare them in a meaningful way.\r\n\r\n# 3. Feature extraction\r\n\r\nThe next step was to extract the features from the images. In other words find things in the pictures that would be unique enough to make a difference between cats and dogs but broad enough so that all dogs would fall into one category and all cats into the other.\r\n\r\n## 3.1. Straight lines\r\n\r\nThe first thing which came to mind was counting and doing other stuff with straight lines in the image.\r\n\r\n### 3.1.1 Canny edge detector\r\n\r\nI used an edge detector algorithm called Canny to preprocess the images which - as the name implies - finds edges in images. Because of my preparation with Photoshop it was quite easy for it to find them. It is not easy to see this step with my drawings, so here is a picture of how it looks like when you do this with a photo instead:\r\n\r\n![Canny on a photo from Wikipedia](https://jeena.net/images/2013/catdog/canny.jpg)\r\n\r\nIt basically removes noise with a gausian filter and then finds the intentisty gradians of the image with help of some trigonometry.\r\n\r\nI did not implement the algorithm myself, instead I used the often used [OpenCV implementation](http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/canny_detector/canny_detector.html).\r\n\r\n### 3.1.2 Hough transform\r\n\r\nTo find the lines I used the [Hough transform](https://en.wikipedia.org/wiki/Hough_transform) algorithm. The red lines are those which the Hough transform algorithm found in the example picture:\r\n\r\n![Hough lines](https://jeena.net/images/2013/catdog/hough.png)\r\n\r\nIt essentially groups edges, which can be imperfect, to object candidates by performing an explicit voting procedure. Detecting straight lines can be done by describing them as y = mx + b where m is the slope of the line and b is the intercept. The line is not represented by descrete points (x1,y1)(x2,y2) but instead as a point(x,y) in the parameter space, which makes detection of lines, which are a bit off, possible. In practice it is still more complicated, please read the [Wikipedia article](https://en.wikipedia.org/wiki/Hough_transform) about it.\r\n\r\nI did not implement it myself but used the often used and tested probabilistic [OpenCV implementation](http://docs.opencv.org/modules/imgproc/doc/feature_detection.html?highlight=houghlinesp#houghlinesp).\r\n\r\n## 3.2. Line features\r\n\r\nI extracted these features from the lines:\r\n\r\n- amount of lines\r\n- average length of lines\r\n- average angle of lines\r\n\r\n## 3.3. Other features\r\n\r\nI also extracted the amount of black pixels in the image to use it as a possible feature which was not using the extracted lines.\r\n\r\n# 4. _k_-nearest neighbor algorithm\r\n\r\nI chose to use the _k_-Nearest Neighbors algorithm which only locally looks at the neighbors of the document in a radius predefined by the user. It assumes that the document is of the same category as the highest number of neighbors within this radius.\r\nIn the following figure you can see that depending if the user choses k = 3, as shown by the solid line, the algorithm will conclude that the document in the center (green smiley) is of the type triangle because most of this three neighbors are triangles. If on the other hand the user choses k = 7, as shown by the dotted line, then the amount of neighbors which are rectangles is greater as the amount of neighbors which are triangles, so it concludes that the smiley is of type rectangle.\r\n\r\n![k-Nearest Neighbours as a graphic](https://jeena.net/images/2013/catdog/k-nearest-neighbours.png)\r\n\r\nIn the picture above you see how it would look with two dimensions. I have been using four features so the algorithm had to check the distance to the neighbours in four dimensions. This is not really more difficult, it is just more to calculate.\r\n\r\n# 5. Results\r\n\r\nThe results were quite encouraging, I assume it is because I only used one style to draw the dogs and one style to draw the cats.\r\n\r\n## 5.1. k-fold Cross-validation\r\n\r\nI used 10 fold cross-validation for every test I did, which means that I used 90% of the available data for the learning algorithms and then the remaining 10% to test how they performed. I repeated this ten times until all data had been used for testing once.\r\n\r\n## 5.2. Results with all features\r\n\r\nWhen I used all of the features and three nearest neighbours I got amazing 100% accuracy, which was kind of suspect because that normally means that you most probably did something wrong.\r\n\r\n## 5.3. Results with a reduced feature set\r\n\r\nTherefore I tried to reduce the features to check if it would perform worse.\r\n\r\n1. When I removed the information about the amount of black pixels basically nothing happened.\r\n2. When I removed the information about the amount of lines and average length at least I got a couple of wrong categorized images, the accuracy went down to 95%.\r\n3. When I removed the information about the average angle of the lines, that was when I got significant errors. The accuracy dropped down to about 60%, which is still better then pure chance.\r\n\r\nSo it seems like the best feature to detect cat and dog face drawings done by me was the average angle of the straight lines in the image.\r\n\r\n# 6. Future study\r\n\r\nThe most important next step would be to gather many more drawings done by other people who use other styles to draw cat and dog faces.\r\n\r\nThen it would be interesting to use other learning algorithms like Bayes, Perceptron, etc.\r\n\r\nAnd then it would be interesting to use this approach on photos of real cats and dogs.\r\n\r\n# 7. Code\r\n```python\r\n#!/usr/bin/env python\r\n\r\nimport cv2, cv, sys, math, os, numpy\r\nfrom scipy.spatial import KDTree\r\n\r\ndef extractFeatures(label):\r\n\r\n\tdirectory = \"img/\" + label + \"/\"\r\n\r\n\tfeatures = []\r\n\r\n\tfor fn in os.listdir(directory):\r\n\r\n\t\timg = cv2.imread(directory + fn, 0)\r\n\r\n\t\t# find edges\r\n\t\tcanny = cv2.Canny(img, 50, 100)\r\n\r\n\t\t# find colored\r\n\t\tblack_pixels = numpy.count_nonzero(img)\r\n\r\n\t\t# find lines lines\r\n\t\tlines = cv2.HoughLinesP(canny, 1, math.pi/360, 5, None, 10, 1)\r\n\r\n\t\tlengths = []\r\n\t\tangles = []\r\n\t\ttry:\r\n\t\t\tfor line in lines[0]:\r\n\t\t\t\tx1, y1, x2, y2 = line\r\n\r\n\t\t\t\t# Pythagoras\r\n\t\t\t\ta2 = math.pow((x1-x2), 2)\r\n\t\t\t\tb2 = math.pow((y1-y2), 2)\r\n\t\t\t\tlength = int(math.sqrt(a2 + b2))\r\n\t\t\t\tlengths.append(length)\r\n\r\n\t\t\t\tangle = int(math.degrees(math.atan((y1-y2) / (x1-x2))))\r\n\t\t\t\tangles.append(angle)\r\n\t\texcept:\r\n\t\t\tpass\r\n\r\n\t\t# print out everything\r\n\t\tlines_count = len(lengths)\r\n\t\tmid_length = sum(lengths) / lines_count\r\n\t\tmid_angle = sum(angles) / lines_count\r\n\r\n\t\tfeatures.append([\r\n\t\t\t[lines_count, mid_length, mid_angle, black_pixels],\r\n\t\t\tlabel\r\n\t\t])\r\n\r\n\treturn features\r\n\r\n\r\nif __name__ == \"__main__\":\r\n\tcats = extractFeatures(\"cat\")\r\n\tdogs = extractFeatures(\"dog\")\r\n\r\n\ttest_count = 5\r\n\r\n\ttest_data = dogs[:test_count] + cats[:test_count] \r\n\ttest_labels = map(lambda a: a[1], test_data)\r\n\ttest_features = map(lambda a: a[0], test_data)\r\n\r\n\tdata = cats[test_count:] + dogs[test_count:]\r\n\tlabels = map(lambda a: a[1], data)\r\n\tfeatures = map(lambda a: a[0], data)\r\n\r\n\ttree = KDTree(features)\r\n\t\r\n\tfor t in xrange(0, test_count * 2):\r\n\t\td, i = tree.query(test_features[t], k=3)\r\n\t\tprint \"-\"\r\n\t\tfor j in xrange(0, len(i)):\r\n\t\t\tprint test_labels[t] + \" is a \" + labels[i[j]]\r\n```\r\n","google":"","note":"Don't delete this file! It's used internally to help with page regeneration."} \ No newline at end of file diff --git a/stylesheets/pygment_trac.css b/stylesheets/pygment_trac.css new file mode 100644 index 0000000..e65cedf --- /dev/null +++ b/stylesheets/pygment_trac.css @@ -0,0 +1,70 @@ +.highlight .hll { background-color: #ffffcc } +.highlight { background: #f0f3f3; } +.highlight .c { color: #0099FF; font-style: italic } /* Comment */ +.highlight .err { color: #AA0000; background-color: #FFAAAA } /* Error */ +.highlight .k { color: #006699; font-weight: bold } /* Keyword */ +.highlight .o { color: #555555 } /* Operator */ +.highlight .cm { color: #0099FF; font-style: italic } /* Comment.Multiline */ +.highlight .cp { color: #009999 } /* Comment.Preproc */ +.highlight .c1 { color: #0099FF; font-style: italic } /* Comment.Single */ +.highlight .cs { color: #0099FF; font-weight: bold; font-style: italic } /* Comment.Special */ +.highlight .gd { background-color: #FFCCCC; border: 1px solid #CC0000 } /* Generic.Deleted */ +.highlight .ge { font-style: italic } /* Generic.Emph */ +.highlight .gr { color: #FF0000 } /* Generic.Error */ +.highlight .gh { color: #003300; font-weight: bold } /* Generic.Heading */ +.highlight .gi { background-color: #CCFFCC; border: 1px solid #00CC00 } /* Generic.Inserted */ +.highlight .go { color: #AAAAAA } /* Generic.Output */ +.highlight .gp { color: #000099; font-weight: bold } /* Generic.Prompt */ +.highlight .gs { font-weight: bold } /* Generic.Strong */ +.highlight .gu { color: #003300; font-weight: bold } /* Generic.Subheading */ +.highlight .gt { color: #99CC66 } /* Generic.Traceback */ +.highlight .kc { color: #006699; font-weight: bold } /* Keyword.Constant */ +.highlight .kd { color: #006699; font-weight: bold } /* Keyword.Declaration */ +.highlight .kn { color: #006699; font-weight: bold } /* Keyword.Namespace */ +.highlight .kp { color: #006699 } /* Keyword.Pseudo */ +.highlight .kr { color: #006699; font-weight: bold } /* Keyword.Reserved */ +.highlight .kt { color: #007788; font-weight: bold } /* Keyword.Type */ +.highlight .m { color: #FF6600 } /* Literal.Number */ +.highlight .s { color: #CC3300 } /* Literal.String */ +.highlight .na { color: #330099 } /* Name.Attribute */ +.highlight .nb { color: #336666 } /* Name.Builtin */ +.highlight .nc { color: #00AA88; font-weight: bold } /* Name.Class */ +.highlight .no { color: #336600 } /* Name.Constant */ +.highlight .nd { color: #9999FF } /* Name.Decorator */ +.highlight .ni { color: #999999; font-weight: bold } /* Name.Entity */ +.highlight .ne { color: #CC0000; font-weight: bold } /* Name.Exception */ +.highlight .nf { color: #CC00FF } /* Name.Function */ +.highlight .nl { color: #9999FF } /* Name.Label */ +.highlight .nn { color: #00CCFF; font-weight: bold } /* Name.Namespace */ +.highlight .nt { color: #330099; font-weight: bold } /* Name.Tag */ +.highlight .nv { color: #003333 } /* Name.Variable */ +.highlight .ow { color: #000000; font-weight: bold } /* Operator.Word */ +.highlight .w { color: #bbbbbb } /* Text.Whitespace */ +.highlight .mf { color: #FF6600 } /* Literal.Number.Float */ +.highlight .mh { color: #FF6600 } /* Literal.Number.Hex */ +.highlight .mi { color: #FF6600 } /* Literal.Number.Integer */ +.highlight .mo { color: #FF6600 } /* Literal.Number.Oct */ +.highlight .sb { color: #CC3300 } /* Literal.String.Backtick */ +.highlight .sc { color: #CC3300 } /* Literal.String.Char */ +.highlight .sd { color: #CC3300; font-style: italic } /* Literal.String.Doc */ +.highlight .s2 { color: #CC3300 } /* Literal.String.Double */ +.highlight .se { color: #CC3300; font-weight: bold } /* Literal.String.Escape */ +.highlight .sh { color: #CC3300 } /* Literal.String.Heredoc */ +.highlight .si { color: #AA0000 } /* Literal.String.Interpol */ +.highlight .sx { color: #CC3300 } /* Literal.String.Other */ +.highlight .sr { color: #33AAAA } /* Literal.String.Regex */ +.highlight .s1 { color: #CC3300 } /* Literal.String.Single */ +.highlight .ss { color: #FFCC33 } /* Literal.String.Symbol */ +.highlight .bp { color: #336666 } /* Name.Builtin.Pseudo */ +.highlight .vc { color: #003333 } /* Name.Variable.Class */ +.highlight .vg { color: #003333 } /* Name.Variable.Global */ +.highlight .vi { color: #003333 } /* Name.Variable.Instance */ +.highlight .il { color: #FF6600 } /* Literal.Number.Integer.Long */ + +.type-csharp .highlight .k { color: #0000FF } +.type-csharp .highlight .kt { color: #0000FF } +.type-csharp .highlight .nf { color: #000000; font-weight: normal } +.type-csharp .highlight .nc { color: #2B91AF } +.type-csharp .highlight .nn { color: #000000 } +.type-csharp .highlight .s { color: #A31515 } +.type-csharp .highlight .sc { color: #A31515 } diff --git a/stylesheets/stylesheet.css b/stylesheets/stylesheet.css new file mode 100644 index 0000000..b48c8a8 --- /dev/null +++ b/stylesheets/stylesheet.css @@ -0,0 +1,427 @@ +/******************************************************************************* +Slate Theme for GitHub Pages +by Jason Costello, @jsncostello +*******************************************************************************/ + +@import url(pygment_trac.css); + +/******************************************************************************* +MeyerWeb Reset +*******************************************************************************/ + +html, body, div, span, applet, object, iframe, +h1, h2, h3, h4, h5, h6, p, blockquote, pre, +a, abbr, acronym, address, big, cite, code, +del, dfn, em, img, ins, kbd, q, s, samp, +small, strike, strong, sub, sup, tt, var, +b, u, i, center, +dl, dt, dd, ol, ul, li, +fieldset, form, label, legend, +table, caption, tbody, tfoot, thead, tr, th, td, +article, aside, canvas, details, embed, +figure, figcaption, footer, header, hgroup, +menu, nav, output, ruby, section, summary, +time, mark, audio, video { + margin: 0; + padding: 0; + border: 0; + font: inherit; + vertical-align: baseline; +} + +/* HTML5 display-role reset for older browsers */ +article, aside, details, figcaption, figure, +footer, header, hgroup, menu, nav, section { + display: block; +} + +ol, ul { + list-style: none; +} + +blockquote, q { +} + +table { + border-collapse: collapse; + border-spacing: 0; +} + +/******************************************************************************* +Theme Styles +*******************************************************************************/ + +body { + box-sizing: border-box; + color:#373737; + background: #212121; + font-size: 16px; + font-family: 'Myriad Pro', Calibri, Helvetica, Arial, sans-serif; + line-height: 1.5; + -webkit-font-smoothing: antialiased; +} + +h1, h2, h3, h4, h5, h6 { + margin: 10px 0; + font-weight: 700; + color:#222222; + font-family: 'Lucida Grande', 'Calibri', Helvetica, Arial, sans-serif; + letter-spacing: -1px; +} + +h1 { + font-size: 36px; + font-weight: 700; +} + +h2 { + padding-bottom: 10px; + font-size: 32px; + background: url('../images/bg_hr.png') repeat-x bottom; +} + +h3 { + font-size: 24px; +} + +h4 { + font-size: 21px; +} + +h5 { + font-size: 18px; +} + +h6 { + font-size: 16px; +} + +p { + margin: 10px 0 15px 0; +} + +footer p { + color: #f2f2f2; +} + +a { + text-decoration: none; + color: #007edf; + text-shadow: none; + + transition: color 0.5s ease; + transition: text-shadow 0.5s ease; + -webkit-transition: color 0.5s ease; + -webkit-transition: text-shadow 0.5s ease; + -moz-transition: color 0.5s ease; + -moz-transition: text-shadow 0.5s ease; + -o-transition: color 0.5s ease; + -o-transition: text-shadow 0.5s ease; + -ms-transition: color 0.5s ease; + -ms-transition: text-shadow 0.5s ease; +} + +#main_content a:hover { + color: #0069ba; + text-shadow: #0090ff 0px 0px 2px; +} + +footer a:hover { + color: #43adff; + text-shadow: #0090ff 0px 0px 2px; +} + +em { + font-style: italic; +} + +strong { + font-weight: bold; +} + +img { + position: relative; + margin: 0 auto; + max-width: 739px; + padding: 5px; + margin: 10px 0 10px 0; + border: 1px solid #ebebeb; + + box-shadow: 0 0 5px #ebebeb; + -webkit-box-shadow: 0 0 5px #ebebeb; + -moz-box-shadow: 0 0 5px #ebebeb; + -o-box-shadow: 0 0 5px #ebebeb; + -ms-box-shadow: 0 0 5px #ebebeb; +} + +pre, code { + width: 100%; + color: #222; + background-color: #fff; + + font-family: Monaco, "Bitstream Vera Sans Mono", "Lucida Console", Terminal, monospace; + font-size: 14px; + + border-radius: 2px; + -moz-border-radius: 2px; + -webkit-border-radius: 2px; + + + +} + +pre { + width: 100%; + padding: 10px; + box-shadow: 0 0 10px rgba(0,0,0,.1); + overflow: auto; +} + +code { + padding: 3px; + margin: 0 3px; + box-shadow: 0 0 10px rgba(0,0,0,.1); +} + +pre code { + display: block; + box-shadow: none; +} + +blockquote { + color: #666; + margin-bottom: 20px; + padding: 0 0 0 20px; + border-left: 3px solid #bbb; +} + +ul, ol, dl { + margin-bottom: 15px +} + +ul li { + list-style: inside; + padding-left: 20px; +} + +ol li { + list-style: decimal inside; + padding-left: 20px; +} + +dl dt { + font-weight: bold; +} + +dl dd { + padding-left: 20px; + font-style: italic; +} + +dl p { + padding-left: 20px; + font-style: italic; +} + +hr { + height: 1px; + margin-bottom: 5px; + border: none; + background: url('../images/bg_hr.png') repeat-x center; +} + +table { + border: 1px solid #373737; + margin-bottom: 20px; + text-align: left; + } + +th { + font-family: 'Lucida Grande', 'Helvetica Neue', Helvetica, Arial, sans-serif; + padding: 10px; + background: #373737; + color: #fff; + } + +td { + padding: 10px; + border: 1px solid #373737; + } + +form { + background: #f2f2f2; + padding: 20px; +} + +img { + width: 100%; + max-width: 100%; +} + +/******************************************************************************* +Full-Width Styles +*******************************************************************************/ + +.outer { + width: 100%; +} + +.inner { + position: relative; + max-width: 640px; + padding: 20px 10px; + margin: 0 auto; +} + +#forkme_banner { + display: block; + position: absolute; + top:0; + right: 10px; + z-index: 10; + padding: 10px 50px 10px 10px; + color: #fff; + background: url('../images/blacktocat.png') #0090ff no-repeat 95% 50%; + font-weight: 700; + box-shadow: 0 0 10px rgba(0,0,0,.5); + border-bottom-left-radius: 2px; + border-bottom-right-radius: 2px; +} + +#header_wrap { + background: #212121; + background: -moz-linear-gradient(top, #373737, #212121); + background: -webkit-linear-gradient(top, #373737, #212121); + background: -ms-linear-gradient(top, #373737, #212121); + background: -o-linear-gradient(top, #373737, #212121); + background: linear-gradient(top, #373737, #212121); +} + +#header_wrap .inner { + padding: 50px 10px 30px 10px; +} + +#project_title { + margin: 0; + color: #fff; + font-size: 42px; + font-weight: 700; + text-shadow: #111 0px 0px 10px; +} + +#project_tagline { + color: #fff; + font-size: 24px; + font-weight: 300; + background: none; + text-shadow: #111 0px 0px 10px; +} + +#downloads { + position: absolute; + width: 210px; + z-index: 10; + bottom: -40px; + right: 0; + height: 70px; + background: url('../images/icon_download.png') no-repeat 0% 90%; +} + +.zip_download_link { + display: block; + float: right; + width: 90px; + height:70px; + text-indent: -5000px; + overflow: hidden; + background: url(../images/sprite_download.png) no-repeat bottom left; +} + +.tar_download_link { + display: block; + float: right; + width: 90px; + height:70px; + text-indent: -5000px; + overflow: hidden; + background: url(../images/sprite_download.png) no-repeat bottom right; + margin-left: 10px; +} + +.zip_download_link:hover { + background: url(../images/sprite_download.png) no-repeat top left; +} + +.tar_download_link:hover { + background: url(../images/sprite_download.png) no-repeat top right; +} + +#main_content_wrap { + background: #f2f2f2; + border-top: 1px solid #111; + border-bottom: 1px solid #111; +} + +#main_content { + padding-top: 40px; +} + +#footer_wrap { + background: #212121; +} + + + +/******************************************************************************* +Small Device Styles +*******************************************************************************/ + +@media screen and (max-width: 480px) { + body { + font-size:14px; + } + + #downloads { + display: none; + } + + .inner { + min-width: 320px; + max-width: 480px; + } + + #project_title { + font-size: 32px; + } + + h1 { + font-size: 28px; + } + + h2 { + font-size: 24px; + } + + h3 { + font-size: 21px; + } + + h4 { + font-size: 18px; + } + + h5 { + font-size: 14px; + } + + h6 { + font-size: 12px; + } + + code, pre { + min-width: 320px; + max-width: 480px; + font-size: 11px; + } + +} diff --git a/tf/README.md b/tf/README.md deleted file mode 100644 index 97e05f8..0000000 --- a/tf/README.md +++ /dev/null @@ -1,3 +0,0 @@ -Categorization of cat and dog faces with TensorFlow -=================================================== - diff --git a/tf/categorize.py b/tf/categorize.py deleted file mode 100755 index 7bb3e96..0000000 --- a/tf/categorize.py +++ /dev/null @@ -1,39 +0,0 @@ -#!/usr/bin/env python3 - -import tensorflow as tf -import pathlib -import random -import matplotlib.pyplot as plt - -tf.enable_eager_execution() - - -data_root = pathlib.Path('../img/') - -all_image_paths = list(data_root.glob('*/*')) -all_image_paths = [str(path) for path in all_image_paths] -random.shuffle(all_image_paths) -image_count = len(all_image_paths) -label_names = sorted(item.name for item in data_root.glob('*/') if item.is_dir()) -label_to_index = dict((name, index) for index,name in enumerate(label_names)) -all_image_labels = [label_to_index[pathlib.Path(path).parent.name] - for path in all_image_paths] - -def preprocess_image(image): - image = tf.image.decode_image(image, channels=3) - image = tf.image.resize_images(image, [192, 192]) - image /= 255.0 # normalize to [0,1] range - return image - -def load_and_preprocess_image(path): - image = tf.read_file(path) - return preprocess_image(image) - -image_path = all_image_paths[0] -label = all_image_labels[0] - -image = load_and_preprocess_image(image_path) -plt.imshow(image) -plt.grid(False) -plt.title(label_names[label].title()) -print()