第二阶段损失值一般到多少才是最好呢

awei669 / VQ-Font

[ICCV 2023] Few shot font generation via transferring similarity guided global and quantization local styles

https://arxiv.org/abs/2309.00827

127 stars 6 forks source link

Open aceliuchanghong opened 1 month ago

aceliuchanghong commented 1 month ago

我训练最多到 L1 loss 0.26左右,效果还是不太像 L1 0.3392 Contrastive 2.5169 D 3.425 G 1.538 B_stl 2.0 B_trg 2.0