U²-Net-small (u2netp), ONNX
An unchanged copy of u2netp.onnx, the small U²-Net (4.6 MB) for finding the main subject of a picture (and so taking its
background out). It is kept here so that UniNotes can download it from a web page, which GitHub's
release files do not allow (they carry no CORS headers).
| file | size | SHA-256 |
|---|---|---|
u2netp.onnx |
4,574,861 bytes | 309c8469258dda742793dce0ebea8e6dd393174f89934733ecc8b14c76f4ddd8 |
Origin and license
- The network and its weights are U²-Net by Xuebin Qin, Zichen Zhang, Chenyang Huang, Masood Dehghan, Osmar Zaiane and
Martin Jagersand, https://github.com/xuebinqin/U-2-Net, Apache License 2.0 (the text is in
LICENSE). - The ONNX file is the one the rembg project (MIT) publishes as
https://github.com/danielgatis/rembg/releases/download/v0.0.0/u2netp.onnx. This copy is byte-for-byte the same; nothing was changed.
If you use it, please cite the paper:
@InProceedings{Qin_2020_PR,
title = {U2-Net: Going Deeper with Nested U-Structure for Salient Object Detection},
author = {Qin, Xuebin and Zhang, Zichen and Huang, Chenyang and Dehghan, Masood and Zaiane, Osmar R. and Jagersand, Martin},
journal = {Pattern Recognition},
volume = {106},
pages = {107404},
year = {2020}
}