pre-processing CelebA

[XavierXiao]

Hi, seems like you use 128 by 128 center crop for celebA, while google use 160 by 160 center crop in their Are GANs Created Equal. See below codes from their repository:

image = tf.image.resize_image_with_crop_or_pad(image, 160, 160) image = tf.image.resize_images(image, [64, 64])

Just wondering if it is a fair comparison? Thanks!

[daib13]

Hi @XavierXiao, thanks for your careful check about the experiment details. We originally used what we believed at the time was one of the most common crop for CelebA, but after digging into the code for the "Are GANs Created Equal? A Large-Scale Study" paper, you are correct that their crop was a bit different than ours (the actual paper itself does not mention the crop). Note that this issue is only related to the CelebA results for the GAN models in Table 1 of our paper "Diagnosing and Enhancing VAE Models," arXiv 2019; the VAE models under default settings in Table 1 all used the same crop. Likewise, in Table 3 all methods apply the same crop as used in the WAE paper "Wasserstein auto-encoders," ICLR 2018.

We are doing more experiments regarding this issue and planning to update the arxiv version with a more clear clarification.

[KomputerMaster64]

@XavierXiao @daib13 Thank you for the discussion. I wantd to know if you have shared the experiments ergarding the pre-processing techniques and various resolutions. Regrads Prabhav