Skip to main content
[Preprint]. 2023 Nov 5:2023.09.27.559646. Originally published 2023 Sep 29. [Version 2] doi: 10.1101/2023.09.27.559646

Figure 2:

Figure 2:

Architecture comparison. Left, compressed NVAE (cNVAE); right, non-hierarchical VAE. We modified the NVAE sampler layer (grey trapezoids) and introduced a deconvolution expand layer (yellow trapezoids). The encoder (inference) and decoder (generation) pathways are depicted in red and blue, respectively. r, residual block; h, trainable parameter; +, feature combination.