Skip to content
/ sige Public

[NeurIPS 2022, T-PAMI 2023] Efficient Spatially Sparse Inference for Conditional GANs and Diffusion Models

License

Notifications You must be signed in to change notification settings

lmxyy/sige

Repository files navigation

Spatially Incremental Generative Engine (SIGE)

[NEW!] SIGE is accepted by T-PAMI!

[NEW!] SIGE supports Stable Diffusion and Mac MPS backend! We also release the codes of an interactive demo for DDPM on M1 Macbook Pro!

[NEW!] SIGE is accepted by NeurIPS 2022! Our code and benchmark datasets are publicly available!

teaser We introduce Spatially Incremental Generative Engine (SIGE),an engine that selectively performs computations at the edited regions for image editing applications. The computation and latency are measured for a single forward. For the above examples, SIGE significantly reduces the computation of SDEdit with DDPM (4-6x) and Stable Diffusion (8x), and GauGAN (15x) while preserving the image quality. When combined with existing model compression methods such as GAN Compression, it further reduces the computation of GauGAN by 47x. On NVIDIA RTX 3090, SIGE achieves up to 7.2x speedups.

Efficient Spatially Sparse Inference for Conditional GANs and Diffusion Models
Muyang Li, Ji Lin, Chenlin Meng, Stefano Ermon, Song Han, and Jun-Yan Zhu
CMU, MIT, and Stanford
In NeurIPS 2022.

Demo

SIGE achieves 2x less conversion time compared to original DDPM on M1 MacBook Pro GPU as we selectively perform computation on the edited regions.

Overview

overview Tiling-based sparse convolution overview. For each convolution Fl in the network, we wrap it into SIGE Convl. The activations of the original image are already pre-computed. When getting the edited image, we first compute a difference mask between the original and edited image and reduce the mask to the active block indices to locate the edited regions. In each SIGE Convl, we directly gather the active blocks from the edited activation Aledited according to the reduced indices, stack the blocks along the batch dimension, and feed them into Fl. The gathered blocks have an overlap of width 2 if Fl is 3×3 convolution with stride 1. After getting the output blocks from Fl, we scatter them back into Fl(Aloriginal) to get the edited output, which approximates Fl(Aledited).

Performance

Efficiency

overview With 1.2% edits, SIGE could reduce the computation of DDPM, Progressive Distillation and GauGAN by 7-18x, achieve a 2-4x speedup on NVIDIA RTX 3090, 3-5x speedup on Apple M1 Pro GPU and 4-14x on M1 Pro CPU. When combined with GAN Compression, it further reduces 50x computation on GauGAN, achieving 38x speedup on M1 Pro CPU. Please check our paper for more details and results.

Quality

overviewQualitative results under different edit sizes. PD is Progressive Distillation. Our method well preserves the visual fidelity of the original model without losing global context.

quality-stable-diffusion

More qualitative results of Stable Diffusion on both image inpainting and editing, measured on NVIDIA RTX 3090.

References:

  • Denoising Diffusion Probabilistic Models (DDPM), Ho et al., ICLR 2020
  • Denoising Diffusion Implicit Model (DDIM), Song et al., ICLR 2021
  • Progressive Distillation for Fast Sampling of Diffusion Models, Salimans et al., ICLR 2022
  • Semantic Image Synthesis with Spatially-Adaptive Normalization (GauGAN), Park et al., CVPR 2019
  • GAN Compression: Efficient Architectures for Interactive Conditional GANs, Li et al., CVPR 2020
  • High-Resolution Image Synthesis with Latent Diffusion Models, Rombach et al., CVPR 2022

Prerequisites

  • Python3
  • CPU, M1 GPU, or NVIDIA GPU + CUDA CuDNN
  • PyTorch >= 1.7. For M1 GPU support, please install PyTorch>=2.0.

Getting Started

Installation

After installing PyTorch, you should be able to install SIGE with PyPI

pip install sige

or via GitHub:

pip install git+https://github.com/lmxyy/sige.git

or locally for development

git clone git@github.com:lmxyy/sige.git
cd sige
pip install -e .

For MPS backend, please set the environment variables:

export PYTORCH_ENABLE_MPS_FALLBACK=1

Usage Example

See example.py for the minimal SIGE convolution example. Please first install SIGE with the above instructions and torchprofile with

pip install torchprofile

Then you can run it with

python example.py

We also have colab example.

Benchmark

To reproduce the results of DDPM and Progressive Distillation or download the LSUN Church editing datasets, please follow the instructions in diffusion/README.md.

To reproduce the results of GauGAN and GAN Compression or download the Cityscapes editing datasets, please follow the instructions in gaugan/README.md.

Citation

If you use this code for your research, please cite our paper.

@inproceedings{li2022efficient,
  title={Efficient Spatially Sparse Inference for Conditional GANs and Diffusion Models},
  author={Li, Muyang and Lin, Ji and Meng, Chenlin and Ermon, Stefano and Han, Song and Zhu, Jun-Yan},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2022}
}

Acknowledgments

Our code is developed based on SDEdit, ddim, diffusion_distillation, gan-compression, dpm-solver, and stable-diffusion. We refer to sbnet for the tiling-based sparse convolution algorithm implementation. Our work is also inspired by the gather/scatter implementations in torchsparse.

We thank torchprofile for MACs measurement, clean-fid for FID computation and drn for Cityscapes mIoU computation.

We thank Yaoyao Ding, Zihao Ye, Lianmin Zheng, Haotian Tang, and Ligeng Zhu for the helpful comments on the engine design. We also thank George Cazenavette, Kangle Deng, Ruihan Gao, Daohan Lu, Sheng-Yu Wang and Bingliang Zhang for their valuable feedback. The project is partly supported by NSF, MIT-IBM Watson AI Lab, Kwai Inc, and Sony Corporation.