ClearPic models (Real-ESRGAN, ONNX)
ONNX exports of the official Real-ESRGAN weights, used by ClearPic to upscale images in the browser with ONNX Runtime Web.
| File | Source weights | Use |
|---|---|---|
realesr-general-x4v3.onnx |
realesr-general-x4v3.pth (v0.2.5.0) |
fast, general photos |
RealESRGAN_x4plus.onnx |
RealESRGAN_x4plus.pth (v0.1.0) |
best quality, photos |
RealESRGAN_x4plus_anime_6B.onnx |
RealESRGAN_x4plus_anime_6B.pth (v0.2.2.4) |
anime / illustrations |
- Input
input: float32,[1, 3, H, W], RGB in[0, 1]; H and W are dynamic. - Output
output: float32,[1, 3, 4H, 4W], RGB in[0, 1](clip it). - Opset 17, exported with
torch.onnx.export; outputs match the PyTorch models to within 1e-5.
License
The weights and the network code are by Xintao Wang et al. and are released under the
BSD 3-Clause License (see LICENSE). These files are a format conversion only.
@InProceedings{wang2021realesrgan,
author = {Xintao Wang and Liangbin Xie and Chao Dong and Ying Shan},
title = {Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data},
booktitle = {International Conference on Computer Vision Workshops (ICCVW)},
date = {2021}
}