|
--- |
|
tags: |
|
- unconditional-image-generation |
|
license: apache-2.0 |
|
--- |
|
|
|
# Model info |
|
|
|
Project [fbanimegan](https://github.com/SkyTNT/fbanimegan) |
|
|
|
### fbanime.pkl |
|
|
|
StyleGan2 model trained with official [StyleGan3](https://github.com/NVlabs/stylegan3). |
|
But I modified the code (networks_stylegan2.py and dataset.py) to support non-square resolutions. |
|
|
|
FID: 1.4 |
|
|
|
### fbanime_fp32.pkl |
|
|
|
fp32 version of fbanime.pkl |
|
|
|
Note: Fp16 version (fbanime.pkl) only works on gpu. And fp32 version works on gpu and cpu. |
|
|
|
### g_mapping.onnx |
|
|
|
onnx format mapping network of fbanime_fp32.pkl |
|
|
|
### g_synthesis.onnx |
|
|
|
onnx format synthesis network of fbanime_fp32.pkl |
|
|
|
### encoder.onnx |
|
|
|
e4e model trained with [encoder4editing-stylegan3](https://github.com/yj7082126/encoder4editing-stylegan3). |
|
I add support for official StyleGan2 model and change backbone to ResNet-34 in [restyle-encoder](https://github.com/yuval-alaluf/restyle-encoder). |
|
|
|
### waifu_dect.onnx |
|
|
|
YOLOv5 model trained with official [YOLOv5](https://github.com/ultralytics/yolov5) |
|
|
|
# Usage |
|
|
|
see [demo](https://huggingface.co/spaces/skytnt/full-body-anime-gan/blob/main/app.py) |
|
|
|
# Dataset |
|
|
|
[fbanimehq](https://huggingface.co/datasets/skytnt/fbanimehq) v2.0 |
|
|
|
|