Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/bes-dev/MobileStyleGAN.pytorch
An official implementation of MobileStyleGAN in PyTorch
https://github.com/bes-dev/MobileStyleGAN.pytorch
gan image-synthesis mobile-development openvino stylegan2 stylegan2-pytorch sylegan
Last synced: 3 months ago
JSON representation
An official implementation of MobileStyleGAN in PyTorch
- Host: GitHub
- URL: https://github.com/bes-dev/MobileStyleGAN.pytorch
- Owner: bes-dev
- License: apache-2.0
- Created: 2021-03-11T23:03:05.000Z (over 3 years ago)
- Default Branch: develop
- Last Pushed: 2022-04-27T22:30:34.000Z (over 2 years ago)
- Last Synced: 2024-06-24T05:54:13.366Z (5 months ago)
- Topics: gan, image-synthesis, mobile-development, openvino, stylegan2, stylegan2-pytorch, sylegan
- Language: Python
- Homepage:
- Size: 318 KB
- Stars: 661
- Watchers: 11
- Forks: 81
- Open Issues: 21
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG
- License: LICENSE
Awesome Lists containing this project
- StarryDivineSky - MobileStyleGAN.pytorch
README
## MobileStyleGAN: A Lightweight Convolutional Neural Network for High-Fidelity Image Synthesis
Official PyTorch Implementation
The accompanying videos can be found on [YouTube](https://www.youtube.com/playlist?list=PLstKhmdpWBtwsvq_27ALmPbf_mBLmk0uI).
For more details, please refer to the [paper](https://arxiv.org/abs/2104.04767).
## Requirements
* Python 3.8+
* 1–8 high-end NVIDIA GPUs with at least 12 GB of memory. We have done all testing and development using DL Workstation with 4x2080Ti
## Training
```bash
pip install -r requirements.txt
python train.py --cfg configs/mobile_stylegan_ffhq.json --gpus
```
## Convert checkpoint from rosinality/stylegan2-pytorch
Our framework supports StyleGAN2 checkpoints format from [rosinality/stylegan2-pytorch](https://github.com/rosinality/stylegan2-pytorch).
To convert ckpt your own checkpoint of StyleGAN2 to our framework:
```bash
python convert_rosinality_ckpt.py --ckpt --ckpt-mnet --ckpt-snet --cfg-path
```
### Check converted checkpoint
To check that your checkpoint is converted correctly, just run demo visualization:
```bash
python demo.py --cfg --ckpt "" --generator teacher
```
## Generate images using MobileStyleGAN
```bash
python generate.py --cfg configs/mobile_stylegan_ffhq.json --device cuda --ckpt --output-path --batch-size --n-batches
```
## Evaluate FID score
To evaluate the FID score we use a modified version of [pytorch-fid](https://github.com/mseitzer/pytorch-fid) library:
```bash
python evaluate_fid.py
```
## Demo
Run demo visualization using MobileStyleGAN:
```bash
python demo.py --cfg configs/mobile_stylegan_ffhq.json --ckpt
```
Run visual comparison using StyleGAN2 vs. MobileStyleGAN:
```bash
python compare.py --cfg configs/mobile_stylegan_ffhq.json --ckpt
```
## Convert to ONNX
```bash
python train.py --cfg configs/mobile_stylegan_ffhq.json --ckpt --export-model onnx --export-dir
```
## Convert to CoreML
```bash
python train.py --cfg configs/mobile_stylegan_ffhq.json --ckpt --export-model coreml --export-dir
```
## Deployment using OpenVINO
We provide external library [random_face](https://github.com/bes-dev/random_face) as an example of deploying our model at the edge devices using the [OpenVINO](https://github.com/openvinotoolkit/openvino) framework.
## Pretrained models
|Name|FID|
|:---|:--|
|[mobilestylegan_ffhq.ckpt](https://drive.google.com/uc?id=11Kja0XGE8liLb6R5slNZjF3j3v_6xydt)|7.75|
(*) Our framework supports automatic download pretrained models, just use `--ckpt `.
## Legacy license
|Code|Source|License|
|:---|:-----|:------|
|[Custom CUDA kernels](core/models/modules/ops/)|https://github.com/NVlabs/stylegan2|[Nvidia License](LICENSE-NVIDIA)|
|[StyleGAN2 blocks](core/models/modules/legacy.py)|https://github.com/rosinality/stylegan2-pytorch|MIT|
## Acknowledgements
We want to thank the people whose works contributed to our project::
* Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, Timo Aila for research related to style based generative models.
* Kim Seonghyeon for implementation of StyleGAN2 in [PyTorch](https://github.com/rosinality/stylegan2-pytorch).
* Fergal Cotter for implementation of Discrete Wavelet Transforms and Inverse Discrete Wavelet Transforms in [PyTorch](https://github.com/fbcotter/pytorch_wavelets).
* Cyril Diagne for the excellent [demo of how to run MobileStyleGAN directly into the web-browser](https://github.com/cyrildiagne/mobilestylegan-web-demo).
## Citation
If you are using the results and code of this work, please cite it as:
```
@misc{belousov2021mobilestylegan,
title={MobileStyleGAN: A Lightweight Convolutional Neural Network for High-Fidelity Image Synthesis},
author={Sergei Belousov},
year={2021},
eprint={2104.04767},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
@article{BELOUSOV2021100115,
title = {MobileStyleGAN.pytorch: PyTorch-based toolkit to compress StyleGAN2 model},
journal = {Software Impacts},
year = {2021},
issn = {2665-9638},
doi = {https://doi.org/10.1016/j.simpa.2021.100115},
url = {https://www.sciencedirect.com/science/article/pii/S2665963821000452},
author = {Sergei Belousov},
}
```