search

akanimax / BMSG-GAN

[MSG-GAN] Any body can GAN! Highly stable and robust architecture. Requires little to no hyperparameter tuning. Pytorch Implementation
MIT License
629 stars 105 forks source link

Data preprocessing tips #19

Closed hbarovertwo closed 5 years ago

hbarovertwo commented 5 years ago

Hello @akanimax , awesome results.

I have my own use case which is generate realistic images of vehicles. I am using a publicly available dataset for testing, I think stanford cars dataset.

I am wondering how important alignment is. For example if I have cars in different orientation (facing left, right etc) should I make them all aligned for the GAN training? If I don't, will this hinder the ability to generate images?

Thanks for any insight, Mason

akanimax commented 5 years ago

@hbarovertwo,

The MSG-GAN technique is quite robust to such variations in the input images. To give you an example, the flowers shown on the README were only resized to 256 x 256 no other kind of cropping / alignment was done. Besides this, I have also trained it on a completely unaligned dataset (open Image dataset) with good results. I also remember @BlindElephants had trained on a dataset extracted from a movie. I would highly encourage you to train without spending too much effort on aligning the dataset.

Hope this helps.

Cheers! @akanimax

hbarovertwo commented 5 years ago

Thanks!! Running an iteration now and will see how it goes!