nachifur / RDDM

CVPR 2024: Residual Denoising Diffusion Models
https://arxiv.org/abs/2308.13712
367 stars 34 forks source link

cumsum_prev 有误? #30

Closed czzbb closed 2 months ago

czzbb commented 2 months ago

您好,请问 https://github.com/nachifur/RDDM/blob/c8a4a846d1f5b22e2c533c601f526ad15980272f/src/residual_denoising_diffusion_pytorch.py#L680 与https://github.com/nachifur/RDDM/blob/c8a4a846d1f5b22e2c533c601f526ad15980272f/src/residual_denoising_diffusion_pytorch.py#L681 行中,pad的value为什么是1?我的理解是0。

nachifur commented 2 months ago

这个填0或1应该影响不大,因为这个“数值”也没有其他含义,仅是对齐维度

czzbb commented 2 months ago

好的,感谢解答~