maskbit
Collection
6 items
•
Updated
This model is the MaskBit tokenizer with a vocabulary size of 18bits. It uses a downsampling factor of 16 and is trained on ImageNet for images of resolution 256.
You can find more details on the project page and in the paper.
Dataset used to train markweber/maskbit_tokenizer_18bit
Collection including markweber/maskbit_tokenizer_18bit
Evaluation results
- rFID on ILSVRC/imagenet-1kself-reported1.160
- InceptionScore on ILSVRC/imagenet-1kself-reported197.800
- LPIPS on ILSVRC/imagenet-1kself-reported0.270
- PSNR on ILSVRC/imagenet-1kself-reported22.000
- SSIM on ILSVRC/imagenet-1kself-reported0.590
- CodebookUsage on ILSVRC/imagenet-1kself-reported0.500