$ρ$-VAE: Autoregressive parametrization of the VAE encoder
This is an incremental improvement for researchers and practitioners using VAEs in image generation, offering a plug-and-play enhancement to model performance.
The authors tackled the problem of improving variational autoencoders (VAEs) for image generation by replacing the standard diagonal Gaussian posterior with a first-order autoregressive Gaussian, which better captures data correlations. This minimal change consistently enhanced image generation quality across all tested setups without requiring additional tuning.
We make a minimal, but very effective alteration to the VAE model. This is about a drop-in replacement for the (sample-dependent) approximate posterior to change it from the standard white Gaussian with diagonal covariance to the first-order autoregressive Gaussian. We argue that this is a more reasonable choice to adopt for natural signals like images, as it does not force the existing correlation in the data to disappear in the posterior. Moreover, it allows more freedom for the approximate posterior to match the true posterior. This allows for the repararametrization trick, as well as the KL-divergence term to still have closed-form expressions, obviating the need for its sample-based estimation. Although providing more freedom to adapt to correlated distributions, our parametrization has even less number of parameters than the diagonal covariance, as it requires only two scalars, $ρ$ and $s$, to characterize correlation and scaling, respectively. As validated by the experiments, our proposition noticeably and consistently improves the quality of image generation in a plug-and-play manner, needing no further parameter tuning, and across all setups. The code to reproduce our experiments is available at \url{https://github.com/sssohrab/rho_VAE/}.