Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42....
Transcript of Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42....
![Page 1: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/1.jpg)
Facial Expression Recognition A.i.
By TEAM BIGAugust 23, 2017
1
![Page 2: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/2.jpg)
Table of Contents
➔ IntroductionWhy we built our A.i.
➔ BodyWhat our A.i can doHow we made our A.i.
➔ ConclusionHow can we improve our A.i.?
2
![Page 3: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/3.jpg)
Introduction
Why did we build Facial Recognition A.i.?
3
![Page 4: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/4.jpg)
우리 프로젝트는?
4
![Page 5: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/5.jpg)
컴퓨터 비전이란…
5
![Page 6: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/6.jpg)
표정분석은 일부분
6
![Page 7: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/7.jpg)
그럼, 왜 표정분석인가?
입력디바이스의진화
표정분석
개발스코프
7
![Page 8: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/8.jpg)
표정분석이 가능해진다면…
8
![Page 9: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/9.jpg)
9
![Page 10: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/10.jpg)
빅데이터란?
10
![Page 11: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/11.jpg)
+ =
11
![Page 12: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/12.jpg)
Body 1
What can our A.i. do?
12
![Page 13: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/13.jpg)
Paul Ekman의 7대 미세 표정
13
![Page 14: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/14.jpg)
회선신경망이란?
•
•
14
![Page 15: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/15.jpg)
15
![Page 16: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/16.jpg)
DEMO
1
16
![Page 18: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/18.jpg)
Demo 1 - ResultsIntensity
Time(sec)
18
![Page 19: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/19.jpg)
DEMO
2
19
![Page 21: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/21.jpg)
21
![Page 22: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/22.jpg)
22
![Page 23: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/23.jpg)
23
![Page 24: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/24.jpg)
24
![Page 25: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/25.jpg)
25
![Page 26: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/26.jpg)
DEMO
직접시연3
26
![Page 27: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/27.jpg)
Demo 1
짧은 영상
Demo 2
영화 전체
Demo 3
직 찍
27
![Page 28: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/28.jpg)
Body 2
How did we build our A.i.?
28
![Page 29: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/29.jpg)
과정
모델 생성 과정
➔ 사진수집 과정
➔ 전처리과정
모델 적용 과정
➔ 영상을 1초별 Frame 자름
➔ Azure API 활용전처리
➔ 자체모델에 입력
➔ CSV Output
➔ R을활용해시각화
29
![Page 30: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/30.jpg)
30
![Page 31: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/31.jpg)
크로울링Python
Google Image Crawler
Keyword
31
![Page 32: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/32.jpg)
Crawling 방법
32
![Page 33: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/33.jpg)
사진 전처리 과정
33
![Page 34: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/34.jpg)
사진 전처리 과정
34
![Page 35: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/35.jpg)
사진 전처리 과정
35
![Page 36: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/36.jpg)
Q.A. - 결측치 기준
36
![Page 37: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/37.jpg)
37
![Page 38: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/38.jpg)
38
![Page 39: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/39.jpg)
39
![Page 40: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/40.jpg)
전처리 결과물
40
![Page 41: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/41.jpg)
전처리 결과물
최대치: 1만6천 장, Neutral
최소치: 266장, Fear
41
![Page 42: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/42.jpg)
모델은 최소치 값으로 제작 (266개)
42
![Page 43: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/43.jpg)
Process 1
전처리:코드
Process 2
전처리:수기
Process 3
클린데이터
266
43
![Page 44: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/44.jpg)
Conclusion
How can we improve our A.i.?
44
![Page 45: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/45.jpg)
10% 향상예측 5%향상예측 5% 향상예측
향후 버전에 대한 로드맵…
45
![Page 46: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/46.jpg)
Reference
•
–
•
–
•
–
•
–
•
–
46
![Page 47: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/47.jpg)
Outside Programs Used
•
•
•
•
•
•
•
•
•
•
47
![Page 48: Facial Expression Recognition A.i.266장, Fear 41. 모델은최소치값으로제작(266개) 42. Process 1 전처리: 코드 Process 2 전처리: 수기 Process 3 클린](https://reader034.fdocuments.net/reader034/viewer/2022050405/5f824b74a39ccc045733b034/html5/thumbnails/48.jpg)
Thank you for watching.Feel free to ask any questions!
48