1)Discriminator uses elu activation
2)Discriminator adopts hinge loss
3)Preprocess places2. Use moblieNet to find the box of various objects, and ensure that mask images don’t cover the box regions.
4)Larger dilated rate
5)256*256 input size when training
6) vgg 19 loss
inpainting key tricks
最新推荐文章于 2024-07-25 21:08:02 发布