迈向可扩散的高维潜空间：频率视角
Toward Diffusible High-Dimensional Latent Spaces: A Frequency Perspective
揭示了高维潜空间中编码器对高频成分欠表达、解码器却强依赖高频潜变量恢复细节的错配，并给出一个无需重训自编码器、可插拔的频率预热课程，可直接套用到任意潜空间扩散/流匹配训练流程中以提升生成质量。
原文依据 decoders depend strongly on high-frequency latent components to recover details, whereas encoders under-represent high-frequency contents
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论潜空间扩散已成为视觉生成的默认范式，然而我们观察到，随着潜空间维度增加，存在一种持续的重建—生成权衡：容量更高的自编码器提升了重建保真度，但生成质量最终会下降。我们将这一差距追溯到高频 token 化与去 token 化过程中的不同行为。通过在 RGB 域与潜空间域进行受控扰动，我们分析了编码器/解码器的行为，发现解码器强烈依赖高频潜变量成分来恢复细节，而编码器对高频内容的表征不足，导致扩散模型训练中高频波段暴露不足并出现欠拟合。为解决该问题，我们提出 FreqWarm，一种即插即用的频率预热课程，在扩散或流匹配训练期间增加早期阶段对高频潜变量信号的暴露——无需修改或重新训练自编码器。在多个高维 tokenizer 上应用后，FreqWarm 稳定提升生成质量：在 Wan2.2-VAE 上使 gFID 降低 14.11，在 LTX-VAE 上降低 6.14，在 DC-AE-f32 上降低 4.42，同时保持架构无关性并兼容多种骨干网络。我们的研究表明，显式管理频率暴露可以成功地将高维潜空间转变为更易扩散的目标。
英文原文摘要 1387 字符
Latent diffusion has become the default paradigm for visual generation, yet we observe a persistent reconstruction-generation trade-off as latent dimensionality increases: higher-capacity autoencoders improve reconstruction fidelity but generation quality eventually declines. We trace this gap to the different behaviors in high-frequency tokenization and detokenization. Through controlled perturbations in both RGB and latent domains, we analyze encoder/decoder behaviors and find that decoders depend strongly on high-frequency latent components to recover details, whereas encoders under-represent high-frequency contents, yielding insufficient exposure and underfitting in high-frequency bands for diffusion model training. To address this issue, we introduce FreqWarm, a plug-and-play frequency warm-up curriculum that increases early-stage exposure to high-frequency latent signals during diffusion or flow-matching training -- without modifying or retraining the autoencoder. Applied across several high-dimensional tokenizers, FreqWarm consistently improves generation quality: decreasing gFID by 14.11 on Wan2.2-VAE, 6.14 on LTX-VAE, and 4.42 on DC-AE-f32, while remaining architecture-agnostic and compatible with diverse backbones. Our study shows that explicitly managing frequency exposure can successfully turn high-dimensional latent spaces into more diffusible targets.