This model is the Taming VQGAN tokenizer with a vocabulary size of 10bits converted into a format for the MaskBit codebase. It uses a downsampling factor of 16 and is trained on ImageNet for images of resolution 256.

You can find more details on the VQGAN in the original repository or paper. All credits for this model belong to Patrick Esser, Robin Rombach and Björn Ommer.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train markweber/taming_vqgan

Paper for markweber/taming_vqgan

Taming Transformers for High-Resolution Image Synthesis

Paper • 2012.09841 • Published Dec 17, 2020

Evaluation results

rFID on ILSVRC/imagenet-1k
self-reported

7.960
InceptionScore on ILSVRC/imagenet-1k
self-reported

115.900
LPIPS on ILSVRC/imagenet-1k
self-reported

0.306
PSNR on ILSVRC/imagenet-1k
self-reported

20.200
SSIM on ILSVRC/imagenet-1k
self-reported

0.520
CodebookUsage on ILSVRC/imagenet-1k
self-reported

0.445