Closed Johnson-yue closed 4 years ago
@Johnson-yue, It's a quite commonly used dataset called Oxford 102 Flowers dataset. This is the link to the dataset -> https://www.robots.ox.ac.uk/~vgg/data/flowers/102/ Hope this helps :smile:
Cheers :beers:! @akanimax
I do not have so many GPU, so, I only can use small dataset to test (I only have one RTX 2080ti),thank you, I will train it with your network
Wow! RTX. I have not had the chance to get a hands on it yet :smile:. Please let me know how you find the performance. BTW, I am soon going to release my MSG-StyleGAN code. Stay tuned. Perhaps you could give that one a try. It has really nice performance. These are my latest flowers results. :smile:
Hope this helps.
Cheers :beers:! @akanimax
Hi, I have downloaded the flower-102 dataset, and it only have 8189 pictures。 something I missing?? or your network can be train with only 8k images and get so good performance?
If I want to training flower dataset , Would you supply a train script ? like using eql or not? using ema or not ? Although I think them should be used, anyway
Oh yeah, the dataset does have only 8K images. The MSG-GAN method works quite well on datasets of small sizes too. We obtained a record FID of 19.6082 with the latest method. But this version also has a score of around ~30 I guess.
Wow, the new method is MSG-styleGAN? brilliant! but now , I think test this version first,how many epoch will be train with flowers?
500 epoch train result:
128x128 with flowers-102 dataset , very nice !! I will test other dataset , thank you
Hi, Would supply the flower dataset link ??