Generative Models of Visually Grounded Imagination

TL;DR

This paper introduces a novel variational auto-encoder framework for visually grounded imagination, enabling the generation of images from abstract concepts with new training objectives and inference methods, evaluated on MNIST-with-attributes and CelebA datasets.

Contribution

It presents a new training objective and product-of-experts inference network for VAEs to generate images from partially specified concepts, advancing visual imagination modeling.

Findings

Proposed method outperforms existing joint image-attribute VAEs.

Introduced MNIST-with-attributes dataset for evaluation.

Achieved better correctness, coverage, and compositionality metrics.

Abstract

It is easy for people to imagine what a man with pink hair looks like, even if they have never seen such a person before. We call the ability to create images of novel semantic concepts visually grounded imagination. In this paper, we show how we can modify variational auto-encoders to perform this task. Our method uses a novel training objective, and a novel product-of-experts inference network, which can handle partially specified (abstract) concepts in a principled and efficient way. We also propose a set of easy-to-compute evaluation metrics that capture our intuitive notions of what it means to have good visual imagination, namely correctness, coverage, and compositionality (the 3 C's). Finally, we perform a detailed comparison of our method with two existing joint image-attribute VAE methods (the JMVAE method of Suzuki et.al. and the BiVCCA method of Wang et.al.) by applying them to two datasets: the MNIST-with-attributes dataset (which we introduce here), and the CelebA dataset.