Reproduce paint to image task - Githubissues

tamarott / SinGAN

Official pytorch implementation of the paper: "SinGAN: Learning a Generative Model from a Single Natural Image"

https://tamarott.github.io/SinGAN.htm

Other

3.29k stars 608 forks source link

Reproduce paint to image task #149

Closed juliecious closed 3 years ago

juliecious commented 3 years ago

Hi Tamar,

Thank you for the great paper and the code.

I was trying to reproduce the paint to image task with the cows image. I have followed the instruction in the readme and the hints in the supplementary document. Somehow my result looks far off from the result in the paper. Could you please kindly have a look at my colab? Have I missed anything?

Thank you.

Also I would like to point out that the description here might not be correct： https://github.com/tamarott/SinGAN/blob/1b4906b9afbd30bf32ea3d3f29ca6c5f4a2a223c/config.py#L30

juliecious commented 3 years ago

I have removed the max_size while training, and change the start scale. Now I have a pretty satisfying result, I guess.

Screenshot 2021-04-21 at 13 40 38

Another attempt with spongebob

Screenshot 2021-04-21 at 13 43 19

👯