|
发表于 2024-9-28 07:50:31
|
显示全部楼层
本帖最后由 pasanonic 于 2024-9-28 08:03 编辑
你是练 SRC 还是 DST
我现在分开用不同的遮罩, src保留脸部所有细节,除了头发(现在大多数遮罩都搞成通用了,SRC的嘴部遮罩默认部分也被抠掉了)
https://dfldata.cc/forum.php?mod ... 3198&highlight=xseg 这个1.0的src用,别看迭代不多(bs16练的),精度不错,我特意写入了 4W多张 SRC图,口腔都没有抠,发丝和局部的一些轮廓要自己改改
== Model name: XSeg ==
== ==
== Current iteration: 1338883 ==
== ==
==-------------------- Model Options --------------------==
== ==
== face_type: wf ==
== batch_size: 16 ==
== pretrain: False ==
== ==
==--------------------- Running On ----------------------==
== ==
== Device index: 0 ==
== Name: NVIDIA GeForce RTX 3070 Laptop GPU ==
== VRAM: 5.34GB ==
== ==
|
|