Each component below is used under its own licence. They run in your browser and are downloaded from editspec.com. Where we changed a component, we say how.
AI models
MODNet: people background removal
- Authors: Zhanghan Ke, Jiayu Sun, Kaican Li, Qiong Yan and Rynson W. H. Lau.
- Licence: Apache License 2.0 (opens in a new tab).
- Source: github.com/ZHKKKe/MODNet (opens in a new tab).
- Changes: we use the official ONNX export of the photographic portrait matting model, with its weights stored as 16-bit floating point numbers to halve the download. Calculations still run at full 32-bit precision.
U²-Net: product cutouts
- Authors: Xuebin Qin, Zichen Zhang, Chenyang Huang, Masood Dehghan, Osmar R. Zaiane and Martin Jagersand.
- Licence: Apache License 2.0 (opens in a new tab).
- Source: github.com/xuebinqin/U-2-Net (opens in a new tab).
- Changes: we use the published u2net weights in ONNX format, reduced to 8 bits with ONNX Runtime’s dynamic quantisation, and split into two files that your browser joins again.
Software
ONNX Runtime Web 1.30.0: runs the AI models
- Copyright: Microsoft Corporation.
- Licence: MIT License (opens in a new tab).
- Source: github.com/microsoft/onnxruntime (opens in a new tab).
- Changes: none.
libheif 1.19.8: opens iPhone HEIC photos
- Copyright: struktur AG and contributors. The build we use, from the libheif-js project, also includes the libde265 video decoder.
- Licence: GNU Lesser General Public License 3.0 (opens in a new tab).
- Source: libheif 1.19.8 (opens in a new tab), libde265 (opens in a new tab) and libheif-js (opens in a new tab) (the exact build we use is libheif-js 1.19.8 (opens in a new tab)).
- Changes: none. It is loaded as separate files (
libheif.jsandlibheif.wasm), so it can be replaced with another build of the same library.
Questions
For a copy of any of these licences or source code, email hello.editspec@gmail.com.