Ken's blog

SMK王将戦予想

2011-09-03T17:10:00.000-07:00

予想を忘れていたので、後だしジャンケンで予想だけ。
敬称略
上から本命→対抗→穴で。

A組：
sky_eagle
ygorou
shoji0113

B組：
ryoumax
GETAYA
taka-c

C組：
sonicspeed
mm.satou
Mackey Naya

D組：
mmmwww
Ton-ton2
yamatetsu814

macでのphpの環境設定

2011-08-24T02:14:00.000-07:00

aptana studio 3というエディタをダウンロードしてみたんだが、設定方法がわからねぇ。
いろいろなサイトを見てるんだけど、その通りにやっても書いてある通りにはならんぞ。
・・・ということで、しばらくは悪戦苦闘になりそうです。

ちなみに、MAMPというのをインストールしてローカルに仮想サーバーを立てた方が早い。
Macには、アパッチが最初からついているからMAMPをインストールしないでもいいのだが、アパッチの設定が面倒だから皆さんがMacでPHPを勉強したいならMAMPのインストールを勧める。

カレンダー機能

2011-08-13T00:02:00.000-07:00

iPhoneアプリ開発をやっているが、IOSにはカレンダー機能がない。

そこで、検索したらオープンソースで配布されている、tapkuLibraryというのがある。

かなり綺麗なカレンダーを作ってくれるし、商用に利用可能とのこと。

SMK棋聖戦予想

2011-07-30T01:38:00.000-07:00

久しぶりなんで、たまには根拠のない予想でもしてみようかなと・・・。

従来の通り山を左上→左下→右上→右下をAからDに分けて予想する。

それでは、予想開始。（敬称略）

A組

本命：fujitsu

対抗：shouji0113

期待：taka-c

本命は、富士さん。自分自身の将棋をちょっと変えようと四苦八苦していた様に感じた。あれから、時間も経っているし、最有力だろう。対抗には、安定性でshouji0113さん。穴には、そろそろ大舞台に出てきそうなたかしーさんを。

B組

本命：nyannkichi

対抗：hamamoto

期待：ma-sa3

本命は、この山で一つ抜けていそうなnyankichiさん。対抗には、SMKの棋戦では毎回良い所まで行っている気がするhamamotoさん。穴には、ma-sa3さん。フジテレビの高橋アナが・・・（略）。

C組

本命：GETAYA

対抗：taku04042005

期待：deck14

本命には、SMKでもトップクラスの下駄屋さん。本人によると、下駄屋ではないらしい・・・。対抗には、何かやってくれそうなtakuさん。穴には、過去対戦して負けそうになったdeck14さん。

D組

本命：ygorou

対抗：sonicspeed

期待：shinichi-t

本命には、調子よければ手合違いの強さを発揮するygorouさん。対抗には、今回はゴキゲン以外にも新しい指し方を見せてくれそうなsonicspeedさん。期待は、私が負けたら全力で応援するshinichi-tさん。まぁ、この組予想はハズレる様に頑張ります。（笑）

美濃崩し200

2011-07-17T21:30:00.000-07:00

美濃崩し200を買って、読んでいるけど、改めて自分が美濃を分かっていないと思った。

回答みなければ、半分も正解はできないと思う。

暗記するくらい繰り返せば、終盤力は上がるんだろうか。

テスト再開

2011-05-09T08:02:00.000-07:00

テストです。

Improved a bit

2009-09-10T18:15:00.001-07:00

2D face image is reconstructed by AAM tool kit. (617 samples)

Myself: Left-142 params (285 samples), Right-207 params (617 samples)

Robert Pattinson: Left-142 params (285 samples), Right-207 params (617 samples)

Barack Obama: Left-142 params (285 samples), Right-207 params (617 samples)

Marilyn Monroe: Left-142 params (285 samples), Right-207 params (617 samples)

Kristin Kreuk: Left-142 params (285 samples), Right-207 params (617 samples)

Harry Potter: Left-142 params (285 samples), Right-207 params (617 samples)

Britney Spears: Left-142 params (285 samples), Right-207 params (617 samples)

Brad Pitt: Left-142 params (285 samples), Right-207 params (617 samples)

Bar Refaeli: Left-142 params (285 samples), Right-207 params (617 samples)

2009-07-09T21:57:00.000-07:00

Left faces are with re-training.
Right faces are without re-training.

2009-07-07T18:11:00.000-07:00

Robert Pattinson
Marilyn Monroe
Kristin Kreuk

Myself
Harry Potter (daniel something?)

Britney Spears

Brad Pitt

Bar Refaeli
Barack Obama

Constructing male faces are not too bad, but female face model looks like a dude!
I think because there are significantly more male samles in my training dataset.

2009-07-02T20:23:00.001-07:00

Top: my original image.
Middile: constructed my 3D face.
Bottom: constructed Obama's 3D face. Original image posted previously.

I re-trained AAM data including 9 test face images. I got 108 numbers after I re-trained.

My face and Obama's face are not too bad.

Result using 99 param

2009-06-29T20:56:00.000-07:00

Top face supposed to be my face.
Bottom face is Obama.

Obama isn't too bad.

result

2009-06-26T17:03:00.001-07:00

Top pic is an image I use. Originally, size of that pic is 100 by 133. I use photoshop to change campus size to 640 by 480. ( try not stretch image itself ).

Middle one is I did not train AAM. I just get parameters after I updated landmarks.

Bottom one is I re-trained AAM, and got parameters after re-training.

I used AAM search to see if it could find Obama face, but it failed in both cases. ( re-train or not ).

face

2009-03-15T13:58:00.001-07:00

I had 100 eigenvectors and reduced to 90.

alpha = 1.

adjusted data = original data matrix - mean

param = (reduced eigenvector)' * adjusted data

reduced face data = reduced eigenvector * param

top: average - (alpha * reduced face data)

middle: average

bottom: average + (alpha * reduced face data)

Things to do during spring break

2009-03-10T10:15:00.000-07:00

1. try to reduce eigenvectors and see how much I can throw eigenvectors away without destroying morphable model.

2. try to create caricture using reduced MM.
parameter = eigenvector * particular face
model = average + parameter * eigenvector

3. think about using neural network to evaluate parameters.

Here is sample of my thesis.
Only abstruct and will modify due to taking neural network idea into this thesis.
http://student.uncw.edu/sh0110/main.pdf

eigenface

2009-01-27T14:10:00.001-08:00

use first 5 eigen vectors.

on top, add eigenvector to average. It's like S(+ + + + +).

on bottom, subtract eigenvector to average. It's like S(- - - - -).

2009-01-15T21:37:00.000-08:00

on top, add x value, subtract y value, and add z value to avarage face. ( +x, -y, +z )

it looks like weight loss.

on bottom, this is average face.

note: first eigenvector is multiplied by 1000.

sample face image

2009-01-14T07:05:00.000-08:00

highest eigen vector times 50 added to average face (right) compare to average face. (left)

so, it's like avg[i][j] + (50 * eigen[i][h])

Because values of emelemts in eigen vector are so small, it will not show huge change unless I multiply by large number such as 1000.

CSC 475 assignment

2008-09-08T06:23:00.000-07:00

1. In making of Jurassic park, they are talking about how to create dinosaur running more realistic. Of course, they use CG to create that scene, but interesting part is how they check dinosaur running. They make dinosaur run backwards to see if dinosaur running looks like realistic.

2. Also, in the scene which actors look a bunch of dinosaurs running, they put tracking light on actors head so that they can place dinosaur on correct place because they know where these actors are looking at.

3. Making dinosaur's sound is interesting part because in the video, they said that we don't actually know the sound of dinosaurs. And, they are right. No one knows the sound of dinosaur crying because we have never saw dinosaurs. However, they combine few sounds to make more scary sound. It may not be actual sound of dinosaur crying, but it sounds good.

other homework can be found: http://student.uncw.edu/sh0110/CSC475.doc

Summary of papers

2008-08-16T21:38:00.000-07:00

I uploaded summary of papers I have read.
http://student.uncw.edu/sh0110/paper.zip

So far, it contains 7 summary papers and original resorce papers.

Reading paper

2008-07-20T20:08:00.000-07:00

Here are papers I have read so far.

http://student.uncw.edu/sh0110/resource.zip

Next step is I have to summarize these paper so that I can add it to thesis paper.

Things that I need to do

2008-05-30T10:48:00.000-07:00

These are things that I need to do by next meeting.

1. USF presentation
2. OpenGL based framework
3. Matrix frameworks
4. Communicate with Fred about design

Small memo
Look for USF database and find format for that.
Think about display image using open source such as OpenGL.

Just create blog

2008-05-20T18:32:00.000-07:00

Blog for my thesis.
Purpose is to keep track of my thesis work so that I can easily figure out where I am at.

Next step is to write title of papers that I have read.

Just for fun.
この日本語が見えるか。(Can you see this Japanese?)