CODEWITHSUNDEEP
tensorflowcomputer-visionconv-neural-networkdeconvolution
Yoni Keren
Yoni Keren

Reputation: 1212

Upscaling in a CNN: Conv Transpose or Tile(Nearest neighbor)?

So far I've seen upscaling in a net using conv transpose (for example in the DCGAN paper).

Now I'm reading a new article by Nvidia (growing GANs- see https://arxiv.org/abs/1710.10196) where they are using tf.tile to "upscale",after which they are using regular convolutions (same padding).

Where can I read more about Nvidia's researchers approach? What is the trade-off between those two approaches?

Upvotes: 1

Views: 1417

Answers (1)

ameroyer
ameroyer

Reputation: 66

I would recommend this Distill article which talks in more details about using "deconvolution" layers vs "upscale + normal convolution".

In particular it seems the latter introduces less artifacts in the generated images.

Upvotes: 2

Related Questions