Skip to Main Content
Table 4: 
Results on image generation on ImageNet- 64 in bits/dim.
ModelLayersHeadsBits/dim
Glow (Kingma and Dhariwal, 2018) − − 3.81 
PixelCNN (Van den Oord et al., 2016) − − 3.57 
PixelSNAIL (Chen et al., 2018) − − 3.52 
SPN (Menick and Kalchbrenner, 2018) − − 3.52 
ImageTransformer (Parmar et al., 2018) 24 16 3.48 
Sparse Transformer (Child et al., 2019) 48 16 3.44 
Reformer (Kitaev et al., 2020) − − 3.65 
 
Routing Transformer 24 16 3.43 
ModelLayersHeadsBits/dim
Glow (Kingma and Dhariwal, 2018) − − 3.81 
PixelCNN (Van den Oord et al., 2016) − − 3.57 
PixelSNAIL (Chen et al., 2018) − − 3.52 
SPN (Menick and Kalchbrenner, 2018) − − 3.52 
ImageTransformer (Parmar et al., 2018) 24 16 3.48 
Sparse Transformer (Child et al., 2019) 48 16 3.44 
Reformer (Kitaev et al., 2020) − − 3.65 
 
Routing Transformer 24 16 3.43 
Close Modal

or Create an Account

Close Modal
Close Modal