NVlabs / DG-Net

:couple: Joint Discriminative and Generative Learning for Person Re-identification. CVPR'19 (Oral) :couple:
https://www.zdzheng.xyz/publication/Joint-di2019
Other
1.27k stars 230 forks source link

DG-Net_Trainer类的forward函数的一个疑问 #64

Closed RuriApoka closed 3 years ago

RuriApoka commented 3 years ago

位于trainer.py文件的第343行, 请问为什么是先self.eval(),然后再生成图片呢?生成完成后再次self.train(), 直接处于默认的train模式会有什么问题吗?

还望指点,谢谢

RuriApoka commented 3 years ago

突然就看明白了,打扰了