Beta-Sigma VAE: Separating beta and decoder variance in Gaussian variational autoencoder

TL;DR

This paper introduces Beta-Sigma VAE, a novel model that explicitly separates beta and decoder variance to improve image synthesis quality and controllability, addressing the blurriness issue in traditional VAEs.

Contribution

The paper proposes Beta-Sigma VAE, which explicitly disentangles beta and decoder variance, enabling better analysis, controllability, and performance in generative modeling.

Findings

Beta-Sigma VAE outperforms conventional VAE in image synthesis.

Explicit separation of beta and variance improves model controllability.

Analysis of rate-distortion curves validates the effectiveness of the approach.

Abstract

Variational autoencoder (VAE) is an established generative model but is notorious for its blurriness. In this work, we investigate the blurry output problem of VAE and resolve it, exploiting the variance of Gaussian decoder and $\beta$ of beta-VAE. Specifically, we reveal that the indistinguishability of decoder variance and $\beta$ hinders appropriate analysis of the model by random likelihood value, and limits performance improvement by omitting the gain from $\beta$. To address the problem, we propose Beta-Sigma VAE (BS-VAE) that explicitly separates $\beta$ and decoder variance $\sigma^2_x$ in the model. Our method demonstrates not only superior performance in natural image synthesis but also controllable parameters and predictable analysis compared to conventional VAE. In our experimental evaluation, we employ the analysis of rate-distortion curve and proxy metrics on computer vision datasets. The code is available on https://github.com/overnap/BS-VAE