DACON
search Subscribe
noti 0

    MY XP

    XP Info

    LEVEL 1

    Until Next Level 150 XP

    0/150 XP

    Today's XP

    0 / 800

    Total XP

    0

    Earned XP

    0

    Spent XP

    0

    Sign in with your SNS accounts

    google
    Sign in with Google
    Don't have an account? Sign Up
    Verify your email

    To sign up, you must verify your email. Do you want to resend the code?
    [Dacon] sign up verification

    Your email must be verified to complete the sign up process. Please verify your email below to complete.

    Terms of Use

    Privacy policy

    Consent to receive marketing information

    The 1st Medical AI (MAI) Competition

    Algorithm | Medical | Gene | Vision | Regression | PCC

    • moneyIcon Prize : 10,000,000 KRW
    • 861 Users Completed
    Practice
    Overview Data Code (share) Talk Leaderboard

     

    ResNet model에 Attention Module(CBAM) 응용 , Regression 최적화 코드

    Thyoon
    2025.02.08 20:13 1,273 Views language
    comments 1

    ResNet model에 Attention Module(CBAM) 응용하여 채널에 관한 피처 어텐션과 spatial(위치) 어텐션 모두를 적용하여 학습성능을 최적화 하였다. 
    로스함수에서는 SmoothL1Loss와 HuberLoss를 결합하여, 피처가 0에 가까워져도, 피처 소실이될 확률을 최소화 하였다. 
    명세에 나와있는 평가지표를 구현은 하였으나 맞는지는 확실치 않아, checkpoint를 선별하는 요소에서 배제하였고, MSE를 주요 평가에 사용하였다.
    피드백은 언제나 환영합니다.

    Code
    1 Comments
    Login Required
    0 / 1000
    Thyoon
    2025.02.12 21:00

    CustomLoss 부분에서 Metric1+Metric2 로 전환 되어야 합니다. 현재 코드로는 HuberLoss만 적용되는 코드입니다.

    📣 The comment input field has been moved to the top of the comment list!
    to_list List
    Previous
    No Previous Post
    Current
    ResNet model에 Attention Module(CBAM) 응용 , Regression 최적화 코드
    Competition - 제1회 Medical AI (MAI) 경진대회
    Likes 10
    Views 1,273
    Comments 1
    7달 전
    Next
    EfficientNet
    Competition - 제1회 Medical AI (MAI) 경진대회
    Likes 5
    Views 1,803
    Comments 1
    10달 전