扩散模型的校正训练
·
0 次浏览
·来源: ArXiv AI
arXiv:2610.14136v2 类型: 新文摘: 在扩散模型中,迭代采样可能会导致误差累积,原因是曝光偏差。降噪器是基于带噪声的真实图像进行训练的,但在推理时,其输入是由其自身之前的不完善预测构建的。
arXiv:2610.14136v2 类型: 新文摘: 在扩散模型中,迭代采样可能会导致误差累积,原因是曝光偏差。降噪器是基于带噪声的真实图像进行训练的,但在推理时,其输入是由其自身之前的不完善预测构建的。
arXiv:2610.14136v2 Announce Type: new Abstract: Iterative sampling in diffusion models can lead to error accumulation due to exposure bias. The denoiser is trained on real images interpolated with noise, but at inference its inputs are constructed from its own imperfect predictions from the earlier ...