|
发表于 2023-3-18 21:44:49
|
显示全部楼层
啊,这个丹超快的啊,迭代一两千次人脸就非常清晰了,专丹你按照我下面这参数炼
== masked_training: True ==
== eyes_mouth_prio: False ==
== uniform_yaw: True ==
== adabelief: True ==
== lr_dropout: n ==
== random_warp: True ==
== true_face_power: 0.01 ==
== face_style_power: 0.0 ==
== bg_style_power: 0.0 ==
== ct_mode: rct ==
== clipgrad: False ==
== pretrain: False ==
== autobackup_hour: 0 ==
== write_preview_history: False ==
== target_iter: 0 ==
== random_src_flip: False ==
== random_dst_flip: True ==
== batch_size: 4 ==
== gan_power: 0.0 ==
== gan_patch_size: 32 ==
== gan_dims: 16 ==
== blur_out_mask: False ==
== random_hsv_power: 0.0 ==
== ==
==------------ Running On -------------==
== ==
== Device index: 0 ==
== Name: Tesla T4 ==
== VRAM: 13.92GB ==
然后训练到人脸轮廓清晰后开启lr,训练到loss下降缓慢,基本是十几万。然后关闭random warp,lr,开启嘴眼一直炼到loss0.1左右,第二列和第一列基本一样可以备份开启gan继续训练了。 |
|