SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations

Authors: Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, Stefano Ermon

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental SDEDIT outperforms state-of-the-art GAN-based methods by up to 98.09% on realism and 91.72% on overall satisfaction scores, according to a human perception study, on multiple tasks, including stroke-based image synthesis and editing as well as image compositing. Section 5: EXPERIMENTS
Researcher Affiliation Academia 1Stanford University 2Carnegie Mellon University
Pseudocode Yes Algorithm 1 Guided image synthesis and editing with SDEdit (VE-SDE); Algorithm 2 Guided image synthesis and editing (VE-SDE); Algorithm 3 Guided image synthesis and editing with mask (VE-SDE); Algorithm 4 Guided image synthesis and editing (VP-SDE); Algorithm 5 Guided image synthesis and editing with mask (VP-SDE)
Open Source Code Yes Our code and models will be available upon publication. (Introduction) Our code is released at https://github.com/ermongroup/SDEdit. (Reproducibility Statement)
Open Datasets Yes We use open source datasets and SDE checkpoints on the corresponding datasets. Based on the availability of open-sourced SDE checkpoints, we use VP-SDE for experiments on LSUN datasets, and VE-SDE for experiments on Celeb A-HQ.
Dataset Splits Yes To quantify realism, we use Kernel Inception Score (KID) between the generated images and the target realistic image dataset (details in Appendix D.2)... KID is computed using the generated images and the corresponding validation sets (see Appendix D.2).
Hardware Specification Yes For scribble-based generation on 256 256 images, SDEdit takes 29.1s to generate one image on one 2080Ti GPU.
Software Dependencies No The paper mentions 'Microsoft Azure Face API' and uses 'pretrained SDE models' from other works, but does not provide specific version numbers for any key software components or libraries.
Experiment Setup Yes In this experiment, we use K = 1, N = 500, t0 = 0.5, for SDEdit (VP). (Appendix D.1) We use t0 = 0.35, N = 700, K = 1 for SDEdit (VE). (Appendix D.1)