issues
search
boost-devs
/
peer-session
🚀 부스트캠프 AI Tech 1기 U-Stage 4조 피어 세션 자료/질문 모음 (archived)
8
stars
2
forks
source link
[DAY 15] Generative model
#60
Closed
opijae
closed
3 years ago
opijae
commented
3 years ago
팀 데이터셋 작업 계획
2월 9일 화요일 14~17시 예정
조유리즈 (김채원, 조유리, 최예나) 구분하는 모델 만들기
한 레이블당 1200장 (train data set : 1000장 test data set 200장)
처음에는 이미지 자르기(crop)이나 전처리 없이 기본으로 해보자
점진적으로 이미지 전처리 , 모델 & 코드 개선으로 성능 업 시키자
질문
후미
(트랜스포머) 멀티헤드에서 사용되는 Wo, 디코더에 들어가는 인풋
https://github.com/boostcamp-ai-tech-4/peer-session/issues/59
히스
generative model 에서 generator 로 실제 아웃풋 생성
https://github.com/boostcamp-ai-tech-4/peer-session/issues/58
펭귄
만 rgb에서 전부 independent 하면 나오는 변수의 개수
(R,G,B) 각각 들어갈 변수의 수가 3개니 '3'이 나온다.
히스
이 식에 대해
다음과 같이 0과 1로 이루어진 그림의 총 픽셀 수 n 일때 dependent일때와 independent일때의 경우의 수
dependent : 서로 다른 픽셀의 값들을 고려함. 각 픽셀에 대한 경우의 수 2, 픽셀의 수 n => 2^n
independent : 서로 다른 픽셀의 값들을 고려하지않고 자신만 생각. 픽셀의 수 n => n
일주일 동안 고생 많으셨습니다.!!
CoodingPenguin
commented
3 years ago
이번 주 모더레이터 정말 수고하셨습니다💪💪
팀 데이터셋 작업 계획
점진적으로 이미지 전처리 , 모델 & 코드 개선으로 성능 업 시키자
질문
후미
(트랜스포머) 멀티헤드에서 사용되는 Wo, 디코더에 들어가는 인풋히스
generative model 에서 generator 로 실제 아웃풋 생성펭귄
만 rgb에서 전부 independent 하면 나오는 변수의 개수히스
이 식에 대해independent : 서로 다른 픽셀의 값들을 고려하지않고 자신만 생각. 픽셀의 수 n => n