I am currently trying to implement DCGAN in keras, but at the beginning of the generator (1024, 4, 4), the kernel of the size (1, 1) is inverted to (512, 8, 8) at stride 5, right?
I'm thinking of using Keras' Conv2 DTranspose, but what should I do with padding the same or valid?Also, is the stride 5 of the kernel okay for the size (4,4)?
If you specify padding="same", the output size will not change, so padding="valid" should be fine.
I can't say for sure (4,4) because I haven't experimented with it, but I don't think strides=5 is possible.In terms of model performance.
© 2024 OneMinuteCode. All rights reserved.