固定VAE方差让流模型训练更简单高效，生成质量超越现有方法。
SimFlow: Simplified and End-to-End Training of Latent Normalizing Flows
- 固定VAE输出方差为常数，简化训练流程
- 在ImageNet 256×256上gFID达2.15，优于STARFlow的2.40
- 可无缝集成REPA-E，gFID进一步降至1.91
归一化流（NFs）通过可逆映射将数据分布与高斯分布对齐。以往方法存在两个问题：一是对训练样本或VAE隐变量添加随机噪声作为数据增强，引入复杂的加噪与去噪步骤；二是使用预训练冻结的VAE编码器，导致重建和生成质量不佳。本文发现，仅将VAE编码器输出的方差固定为常数（如0.5），即可同时解决上述问题。一方面，该策略使编码器输出更广泛的隐变量分布，解码器可从中学习重建干净图像，无需额外设计去噪模块；另一方面，固定方差简化了变分下界，使联合训练NF与VAE更加稳定。在ImageNet $256 \times 256$ 生成任务中，所提模型SimFlow取得2.15的gFID，优于当前最优方法STARFlow（2.40）。此外，SimFlow可无缝集成端到端表示对齐（REPA-E）方法，进一步提升至1.91，成为归一化流类模型的新基准。
原文摘要 · Abstract (English)
Normalizing Flows (NFs) learn invertible mappings between the data and a Gaussian distribution. Prior works usually suffer from two limitations. First, they add random noise to training samples or VAE latents as data augmentation, introducing complex pipelines including extra noising and denoising steps. Second, they use a pretrained and frozen VAE encoder, resulting in suboptimal reconstruction and generation quality. In this paper, we find that the two issues can be solved in a very simple way: just fixing the variance (which would otherwise be predicted by the VAE encoder) to a constant (e.g., 0.5). On the one hand, this method allows the encoder to output a broader distribution of tokens and the decoder to learn to reconstruct clean images from the augmented token distribution, avoiding additional noise or denoising design. On the other hand, fixed variance simplifies the VAE evidence lower bound, making it stable to train an NF with a VAE jointly. On the ImageNet $256 \times 256$ generation task, our model SimFlow obtains a gFID score of 2.15, outperforming the state-of-the-art method STARFlow (gFID 2.40). Moreover, SimFlow can be seamlessly integrated with the end-to-end representation alignment (REPA-E) method and achieves an improved gFID of 1.91, setting a new state of the art among NFs.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。