hanson-young / nniefacelib

nniefacelib是一个在海思35xx系列芯片上运行的人脸算法库
BSD 2-Clause "Simplified" License
554 stars 180 forks source link

为什么对于lds个数为68的数据,训练时只是使用部分点的数据呢。 #76

Closed BinZheng1-pp closed 1 year ago

BinZheng1-pp commented 2 years ago

https://github.com/hanson-young/nniefacelib/blob/d014ffe1868656c0c19866ba79d4d5ca97a4301a/PFPLD/pfld/loss.py#L58

hanson-young commented 2 years ago

因为并不是每个点的标注都和106个点的一致,或许你看起来比较像,但实际训练出来是有差异的

---原始邮件--- 发件人: @.> 发送时间: 2022年6月13日(周一) 上午10:07 收件人: @.>; 抄送: @.***>; 主题: [hanson-young/nniefacelib] 为什么对于lds个数为68的数据,训练时只是使用部分点的数据呢。 (Issue #76)

https://github.com/hanson-young/nniefacelib/blob/d014ffe1868656c0c19866ba79d4d5ca97a4301a/PFPLD/pfld/loss.py#L58

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>

BinZheng1-pp commented 2 years ago

验证了下的确是这样,多谢