About ImageNet Pretraining

[frostinassiky]

Dear MAGVIT authors, thanks for releasing the code of the promising work!

According to the ablation, training from a 2D-VQ model that pre-trained on ImageNet is very essential, but I was not able to find the details of the pre-training process. Would you mind sharing the pre-training configurations, such as batch size, number of epochs, FID score, etc., of the 2D VQ models?