본문 바로가기 메뉴 바로가기

홍석쓰 블로그

프로필사진
  • 글쓰기
  • 관리
  • 태그
  • 방명록
  • RSS

홍석쓰 블로그

검색하기 폼
  • 분류 전체보기 (155)
    • Research (연구 관련) (151)
  • 방명록

pytorch batch normalization (1)
pytorch batch normalization

모듈화된 모델을 사용하고 있는데, 하나의 모듈을 pretrain 시킨 후 전체 모델에 통합시킨 후 다시 학습시켰을 때 성능이 떨어지는 현상이 있었다. Formulation이 아래와 같을 때 Formulation: input -> middle output -> final output input->middle output을 pretrain했다는 것이고, 이 때 middle output의 error가 50이었는데, 전체 모델에 통합시킨 후 다시 학습시켰을 때 middle output의 error가 55로 안 좋아졌다는 말이다. 원인은 2가지가 있는데, 1. pretrain했을 때의 learning rate와 전체 모델에 통합시킨 후에 학습시킬 때의 learning rate가 다르다는 점, 2. pytorch ..

Research (연구 관련) 2020. 2. 5. 15:24
이전 1 다음
이전 다음
공지사항
최근에 올라온 글
최근에 달린 댓글
Total
Today
Yesterday
링크
TAG
  • 에디톨로지
  • Interview
  • focal length
  • nerf
  • pytorch
  • demo
  • deep learning
  • pyrender
  • 컴퓨터비전
  • world coordinate
  • 인터뷰
  • Transformation
  • camera coordinate
  • 머신러닝
  • 비전
  • Machine Learning
  • 피트니스
  • spin
  • 문경식
  • 2d pose
  • densepose
  • Generative model
  • 컴퓨터비젼
  • 헬스
  • part segmentation
  • Docker
  • Pose2Mesh
  • Virtual Camera
  • VAE
  • nohup
more
«   2025/05   »
일 월 화 수 목 금 토
1 2 3
4 5 6 7 8 9 10
11 12 13 14 15 16 17
18 19 20 21 22 23 24
25 26 27 28 29 30 31
글 보관함

Blog is powered by Tistory / Designed by Tistory

티스토리툴바