Unleashing Transformers: Parallel Token Prediction with Discrete Absorbing Diffusion for Fast High-Resolution Image Generation from Vector-Quantized Codes

Bond-Taylor, S.E.; Hessey, P.; Sasaki, H.; Breckon, T.P.; Willcocks, C.G.

Abstract

Whilst diffusion probabilistic models can generate high quality image content, key limitations remain in terms of both generating high-resolution imagery and their associated high computational requirements. Recent Vector-Quantized image models have overcome this limitation of image resolution but are prohibitively slow and unidirectional as they generate tokens via element-wise autoregressive sampling from the prior. By contrast, in this paper we propose a novel discrete diffusion probabilistic model prior which enables parallel prediction of Vector-Quantized tokens by using an unconstrained Transformer architecture as the backbone. During training, tokens are randomly masked in an order-agnostic manner and the Transformer learns to predict the original tokens. This parallelism of Vector-Quantized token prediction in turn facilitates unconditional generation of globally consistent high-resolution and diverse imagery at a fraction of the computational expense. In this manner, we can generate image resolutions exceeding that of the original training set samples whilst additionally provisioning per-image likelihood estimates (in a departure from generative adversarial approaches). Our approach achieves state-of-the-art results in terms of the manifold overlap metrics Coverage (LSUN Bedroom: 0.83; LSUN Churches: 0.73; FFHQ: 0.80) and Density (LSUN Bedroom: 1.51; LSUN Churches: 1.12; FFHQ: 1.20), and performs competitively on FID (LSUN Bedroom: 3.27; LSUN Churches: 4.07; FFHQ: 6.11) whilst offering advantages in terms of both computation and reduced training set requirements.

Citation

Bond-Taylor, S., Hessey, P., Sasaki, H., Breckon, T., & Willcocks, C. (2022). Unleashing Transformers: Parallel Token Prediction with Discrete Absorbing Diffusion for Fast High-Resolution Image Generation from Vector-Quantized Codes. In ECCV 2022: Computer Vision – ECCV 2022 (170-188)

• Conference Name: ECCV 2022: European Conference on Computer Vision
• Conference Location: Tel Aviv, Israel
• Start Date: Oct 23, 2022
• End Date: Oct 27, 2022
• Acceptance Date: Jul 8, 2022
• Online Publication Date: Oct 28, 2022
• Publication Date: 2022-10
• Deposit Date: Oct 12, 2022
• Publicly Available Date: Oct 29, 2023
• Publisher: Springer Verlag
• Volume: 13683
• Pages: 170-188
• Series Title: Lecture Notes in Computer Science
• Book Title: ECCV 2022: Computer Vision – ECCV 2022
• Publisher URL: https://eccv2022.ecva.net/