MacTok: Robust Continuous Tokenization for Image Generation

2026-03-31

AI Summary

Existing continuous image tokenizers suffer from posterior collapse under extremely low token budgets, failing to preserve meaningful visual information. This work proposes MacTok, a 1D continuous tokenizer that integrates stochastic masking with DINO-guided semantic masking. For the first time, it incorporates semantic-aware masking and a global–local representation alignment mechanism into a variational framework, enabling the learning of compact yet robust latent representations from incomplete visual evidence. The approach effectively mitigates posterior collapse, achieving high-fidelity image generation on ImageNet with only 64–128 tokens: it attains a gFID of 1.44 at 256×256 resolution and a state-of-the-art gFID of 1.52 at 512×512, reducing token usage by up to 64× compared to existing methods.

Abstract

Continuous image tokenizers enable efficient visual generation, and those based on variational frameworks can learn smooth, structured latent representations through KL regularization. Yet this often leads to posterior collapse when using fewer tokens, where the encoder fails to encode informative features into the compressed latent space. To address this, we introduce \textbf{MacTok}, a \textbf{M}asked \textbf{A}ugmenting 1D \textbf{C}ontinuous \textbf{Tok}enizer that leverages image masking and representation alignment to prevent collapse while learning compact and robust representations. MacTok applies both random masking to regularize latent learning and DINO-guided semantic masking to emphasize informative regions in images, forcing the model to encode robust semantics from incomplete visual evidence. Combined with global and local representation alignment, MacTok preserves rich discriminative information in a highly compressed 1D latent space, requiring only 64 or 128 tokens. On ImageNet, MacTok achieves a competitive gFID of 1.44 at 256$\times$256 and a state-of-the-art 1.52 at 512$\times$512 with SiT-XL, while reducing token usage by up to 64$\times$. These results confirm that masking and semantic guidance together prevent posterior collapse and achieve efficient, high-fidelity tokenization.

Problem

• posterior collapse
• continuous tokenization
• image generation
• latent representation
• variational frameworks

Innovation

• continuous tokenization
• posterior collapse
• masked augmentation
• semantic guidance
• representation alignment