arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2303.09556cs.CV
收藏

基于最小信噪比加权策略的高效扩散模型训练

Efficient Diffusion Training via Min-SNR Weighting Strategy

Southeast University(东南大学);Microsoft Research Asia(微软亚洲研究院)

Tiankai Hang, Shuyang Gu, Chen Li, Jianmin Bao, Dong Chen, Han Hu, Xin Geng, Baining Guo

AI总结:

针对去噪扩散模型训练收敛慢的问题,提出基于截断信噪比调整时间步损失权重的Min-SNR-γ方法,平衡时间步优化冲突,实现3.4倍加速并刷新ImageNet 256×256基准FID纪录。

摘要:

去噪扩散模型已成为图像生成的主流方法,但这类模型的训练通常存在收敛缓慢的问题。本文发现,收敛缓慢的部分原因在于不同时间步之间的优化方向存在冲突。为解决该问题,我们将扩散训练视为多任务学习问题,提出了一种简单且有效的方法——Min-SNR-γ。该方法基于截断后的信噪比调整各时间步的损失权重,有效平衡了不同时间步之间的优化冲突。实验结果表明,该方法显著提升了收敛速度,比此前的加权策略快3.4倍。同时它的生成效果也更优,在ImageNet 256×256基准上取得了2.06的FID分数新纪录,且所用的模型架构比此前的最优模型更小。代码已发布于https://github.com/TiankaiHang/Min-SNR-Diffusion-Training。

英文摘要:

Denoising diffusion models have been a mainstream approach for image generation, however, training these models often suffers from slow convergence. In this paper, we discovered that the slow convergence is partly due to conflicting optimization directions between timesteps. To address this issue, we treat the diffusion training as a multi-task learning problem, and introduce a simple yet effective approach referred to as Min-SNR-$γ$. This method adapts loss weights of timesteps based on clamped signal-to-noise ratios, which effectively balances the conflicts among timesteps. Our results demonstrate a significant improvement in converging speed, 3.4$\times$ faster than previous weighting strategies. It is also more effective, achieving a new record FID score of 2.06 on the ImageNet $256\times256$ benchmark using smaller architectures than that employed in previous state-of-the-art. The code is available at https://github.com/TiankaiHang/Min-SNR-Diffusion-Training.

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。

↑