Efficient Spatially Sparse Inference for Conditional GANs and Diffusion Models

TL;DR

This paper introduces Spatially Sparse Inference (SSI), a technique that accelerates image editing in generative models by reusing features and selectively computing only edited regions, significantly reducing inference time.

Contribution

The paper proposes SSI and SIGE, enabling efficient, incremental inference for conditional GANs and diffusion models, with extensions for attention layers and support for new hardware.

Findings

SIGE accelerates DDPM by up to 4.6x on GPU

SIGE reduces Stable Diffusion inference time by up to 7.2x

Supports large, sequential edits with significant speedups

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 $1\%$-area edits, SIGE accelerates DDPM by $3.0\times$ on NVIDIA RTX 3090 and $4.6\times$ on Apple M1 Pro GPU, Stable Diffusion by $7.2\times$ on 3090, and GauGAN by $5.6\times$ on 3090 and $5.2\times$ 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.