Exploration into translation-equivariant image quantization

Abstract

Recently, vector-quantized image modeling has demonstrated impressive performance on generation tasks such as text-to-image generation. However, we discover that the current image quantizers do not satisfy translation equivariance in the quantized space due to aliasing. Instead of focusing on anti-aliasing, we propose a simple but effective way to achieve translation-equivariant image quantization by enforcing orthogonality among the codebook embeddings. To explore the advantages of translation-equivariant image quantization, we conduct three experiments with a carefully controlled dataset: (1) text-to-image generation, where the quantized image indices are the target to predict, (2) image-to-text generation, where the quantized image indices are given as a condition, (3) using a smaller training set to analyze sample efficiency. From the strictly controlled experiments, we empirically verify that translation-equivariant image quantizer improves not only sample efficiency but also the accuracy over VQGAN up to +11.9\% in text-to-image generation and +3.9\% in image-to-text generation.