本文介绍: PI(xk)P^I(x^k)PI(xk)=1C∑yik=11ni{1over C}sumlimits_{y_i^k=1}{1over{n_i}}C1​yik​=1∑​ni​1​,PiC(xk)P^C_i(x^k)PiC​(xk)=1C1ni{1over C}{1over{n_i}}C1​ni​1​rik_i^kik​=PiC(xk)PI(xk){P^C_i(x^k)}over{P^I(x^k)}PI(xk)PiC​(xk)​NTR means ‘negativetolerant regu

Distribution-Balanced Loss

在这里插入图片描述
在这里插入图片描述

P

I

(

x

k

)

P^I(x^k)

PI(xk)=

1

C

y

i

k

=

1

1

n

i

{1over C}sumlimits_{y_i^k=1}{1over{n_i}}

C1yik=1ni1

P

i

C

(

x

k

)

P^C_i(x^k)

PiC(xk)=

1

C

1

n

i

{1over C}{1over{n_i}}

C1ni1
r

i

k

_i^k

ik=

P

i

C

(

x

k

)

P

I

(

x

k

)

{P^C_i(x^k)}over{P^I(x^k)}

PI(xk)PiC(xk)
NTR means ‘negative-tolerant regularization’,R-BCE performs a rebalanced weighting strategy

辅助信息

作者使用的’MMDataParallelhas been removed,不建议复现

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注