Which was the train set of DeepFill v1 ImageNet model?

[GiorgioUghini]

Hello, from the readme I see that "all the pretrained models are trained with images of resolution 256x256 and largest hole size 128x128". However, as this work is often used in research papers, I would know which subset of ImageNet you used for training the ImageNet model of DeepFill v1. Can you help me if you remember which set you used? Thanks in advace