arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2301.11706cs.LGcs.AIcs.CV
收藏

输入扰动减少扩散模型中的曝光偏差

Input Perturbation Reduces Exposure Bias in Diffusion Models

Utrecht University(乌得勒支大学);University of Modena and Reggio Emilia(摩德纳-雷焦艾米利亚大学)

Mang Ning, Enver Sangineto, Angelo Porrello, Simone Calderara, Rita Cucchiara

AI总结:

本文提出输入扰动正则化方法,通过模拟推理误差来缓解扩散模型中的曝光偏差,在CelebA上取得SOTA FID 1.27并节省37.5%训练时间。

摘要:

去噪扩散概率模型已展现出令人印象深刻的生成质量,尽管其长采样链导致高计算成本。在本文中,我们观察到长采样链也会导致误差累积现象,这与自回归文本生成中的曝光偏差问题类似。具体来说,我们注意到训练与测试之间存在差异,因为前者以真实样本为条件,而后者以先前生成的结果为条件。为了缓解这一问题,我们提出了一种非常简单但有效的训练正则化方法,即扰动真实样本以模拟推理时的预测误差。我们通过实验表明,在不影响召回率和精确率的情况下,所提出的输入扰动显著提升了样本质量,同时减少了训练和推理时间。例如,在CelebA 64×64上,我们取得了新的最先进FID分数1.27,同时节省了37.5%的训练时间。代码公开于https://github.com/forever208/DDPM-IP。

英文摘要:

Denoising Diffusion Probabilistic Models have shown an impressive generation quality, although their long sampling chain leads to high computational costs. In this paper, we observe that a long sampling chain also leads to an error accumulation phenomenon, which is similar to the exposure bias problem in autoregressive text generation. Specifically, we note that there is a discrepancy between training and testing, since the former is conditioned on the ground truth samples, while the latter is conditioned on the previously generated results. To alleviate this problem, we propose a very simple but effective training regularization, consisting in perturbing the ground truth samples to simulate the inference time prediction errors. We empirically show that, without affecting the recall and precision, the proposed input perturbation leads to a significant improvement in the sample quality while reducing both the training and the inference times. For instance, on CelebA 64$\times$64, we achieve a new state-of-the-art FID score of 1.27, while saving 37.5% of the training time. The code is publicly available at https://github.com/forever208/DDPM-IP

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑