Efficient Spatially Sparse Inference for Conditional GANs and Diffusion Models
Abstract
During image editing, existing deep generative models tend to re-synthesize the entire output from scratch, including the unedited regions. This leads to a significant waste of computation, especially for minor editing operations. In this work, we present Spatially Sparse Inference (SSI), a general-purpose technique that selectively performs computation for edited regions and accelerates various generative models, including both conditional GANs and diffusion models. Our key observation is that users prone to gradually edit the input image. This motivates us to cache and reuse the feature maps of the original image. Given an edited image, we sparsely apply the convolutional filters to the edited regions while reusing the cached features for the unedited areas. Based on our algorithm, we further propose Sparse Incremental Generative Engine (SIGE) to convert the computation reduction to latency reduction on off-the-shelf hardware. With about -area edits, SIGE accelerates DDPM by on NVIDIA RTX 3090 and on Apple M1 Pro GPU, Stable Diffusion by on 3090, and GauGAN by on 3090 and on M1 Pro GPU. Compared to our conference version, we extend SIGE to accommodate attention layers and apply it to Stable Diffusion. Additionally, we offer support for Apple M1 Pro GPU and include more results with large and sequential edits.
Keywords
Cite
@article{arxiv.2211.02048,
title = {Efficient Spatially Sparse Inference for Conditional GANs and Diffusion Models},
author = {Muyang Li and Ji Lin and Chenlin Meng and Stefano Ermon and Song Han and Jun-Yan Zhu},
journal= {arXiv preprint arXiv:2211.02048},
year = {2023}
}
Comments
NeurIPS 2022 T-PAMI 2023 Website: https://www.cs.cmu.edu/~sige/ Code: https://github.com/lmxyy/sige