1cdf14d1e3699d61d237cf76ce1c2dca-Supplemental.pdf

We follow [21] and implement our image compression models as "VQGANs". More specifically, we use the official implementation provided athttps://github.com/CompVis/ We can efficiently model the conditionalsptθ with a sequence-to-sequence model and follow [75, 45] to implement ptθ with an encoder-decoder architecture. Figure 15: Conditionally guided inpainting results obtained from conditional ImageBART trained on the ImageNetdataset. While our approach also models each transition autoregressively from the top-left to the bottomright, each transition additionally has access to global context from the previous step. Besides Figure 1, 6 additional visualizations of this process can be found in Figure 1.