Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

MaskGIT: Masked Generative Image Transformer

About

Generative transformers have experienced rapid popularity growth in the computer vision community in synthesizing high-fidelity and high-resolution images. The best generative transformer models so far, however, still treat an image naively as a sequence of tokens, and decode an image sequentially following the raster scan ordering (i.e. line-by-line). We find this strategy neither optimal nor efficient. This paper proposes a novel image synthesis paradigm using a bidirectional transformer decoder, which we term MaskGIT. During training, MaskGIT learns to predict randomly masked tokens by attending to tokens in all directions. At inference time, the model begins with generating all tokens of an image simultaneously, and then refines the image iteratively conditioned on the previous generation. Our experiments demonstrate that MaskGIT significantly outperforms the state-of-the-art transformer model on the ImageNet dataset, and accelerates autoregressive decoding by up to 64x. Besides, we illustrate that MaskGIT can be easily extended to various image editing tasks, such as inpainting, extrapolation, and image manipulation.

Huiwen Chang, Han Zhang, Lu Jiang, Ce Liu, William T. Freeman• 2022

Related benchmarks

TaskDatasetResultRank
Mathematical ReasoningGSM8K
Accuracy78.8
983
Code GenerationHumanEval--
850
Mathematical ReasoningGSM8K (test)
Accuracy52.2
797
Mathematical ReasoningMATH
Accuracy18
535
Text-to-Image GenerationGenEval
Overall Score52
467
Class-conditional Image GenerationImageNet 256x256
Inception Score (IS)355.6
441
Image GenerationImageNet 256x256 (val)
FID6.18
307
Class-conditional Image GenerationImageNet 256x256 (train)
IS355.6
305
Class-conditional Image GenerationImageNet 256x256 (val)
FID4.92
293
Image GenerationImageNet 256x256
FID4.02
243
Showing 10 of 95 rows
...

Other info

Follow for update