BLACK images generated after changing networks. Is that because of the instability of GAN?

menyifang / ADGAN

The Implementation of paper "Controllable Person Image Synthesis with Attribute-Decomposed GAN" CVPR 2020 (Oral); Pose and Appearance Attributes Transfer;

473 stars 90 forks source link

BLACK images generated after changing networks. Is that because of the instability of GAN? #17

Open Oranging1 opened 3 years ago

Oranging1 commented 3 years ago

When I try to change the network，reducing the number of resnet layers. The results become a BLACK image, which means all the output is nan, Is that because of the instability of GAN? How can I fix it? @menyifang

BadourAlBahar commented 3 years ago

I have the same issue. I, however, am not changing the model. I am using pytorch 1.2.0.