Giters

HJ-harry / MCG_diffusion

Official PyTorch implementation of the NeurIPS 2022 paper "Improving Diffusion Models for Inverse Problems using Manifold Constraints (https://arxiv.org/abs/2206.00941)"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

About freq usage in predictor and corrector algorithm (controllable_generation.py:192)

wenjing11 opened this issue · comments

commented

Thanks for your amazing work!

In controllable_generation.py line 192, variable "freq" is used to select different corrector algorithm, but I do not find any description in MCG paper or Song's paper.
Besides, I do not find trainning code in repo, which is important for me to reproduce the result.

Any help will be appreciated!

commented

You can use the training script directly from this repo and use the trained model for use in the codes provided in our repository. freq is used to control the frequency of corrector steps to be applied per predictor step, but it is always set to 1. You can simply ignore it.