ISNet general-use model (DIS, Dichotomous Image Segmentation)
Copyright (c) Xuebin Qin and the DIS authors.
Licensed under the Apache License, Version 2.0 (see apache-2.0.txt).
Source: https://github.com/xuebinqin/DIS
Paper: Qin et al., "Highly Accurate Dichotomous Image Segmentation", ECCV 2022.

Changes: the original weights were converted to the ONNX format and quantized
to 8 bits so that they can run in a web browser. The files are served as four
chunks in /model/ and are reassembled by the page before use.