xxgege / StableNet

Official repository for CVPR21 paper "Deep Stable Learning for Out-Of-Distribution Generalization".
182 stars 43 forks source link

Shape of weight in RFF #11

Open superpig99 opened 1 year ago

superpig99 commented 1 year ago

Hi, according to the code, specifically Line 40 of loss_reweighting.py, the shape of weight w is (1,1). Does it mean each sample shares the same weight in RFF? image image

maojingen commented 1 year ago

I have also found a similar problem, where all weight values are concentrated around 0.0078. Have you resolved it? need to adjust the parameters in the config ?

Taocunguo commented 2 months ago

我还发现了一个类似的问题,所有权重值都集中在 0.0078 附近。你解决了吗?需要调整配置中的参数吗?

I have also found a similar problem, where all weight values are concentrated around 0.0078. Have you resolved it? need to adjust the parameters in the config ?

Hello, I found it, too. Have you solved it