IS-Net (general use) — the cut-out model BrandTactics runs on your device for Remove background.

Original work: "Highly Accurate Dichotomous Image Segmentation" (IS-Net / DIS), Xuebin Qin, Hang Dai,
Xiaobin Hu, Deng-Ping Fan, Ling Shao, Luc Van Gool — https://github.com/xuebinqin/DIS
Licensed under the Apache License, Version 2.0 (isnet-LICENSE.txt, beside this file).
The ONNX export used: isnet-general-use.onnx as distributed by rembg (https://github.com/danielgatis/rembg,
release v0.0.0), SHA-256 60920e99c45464f2ba57bee2ad08c919a52bbf852739e96947fbb4358c0d964a.

Modified by BrandTactics (BrandMoves), 12 October 2026: converted to fp16 — every float weight of more
than 16 values is stored as float16 and cast back to float32 when the model is loaded; the network and
its answers are otherwise unchanged. The result is isnet-general-use-w16-10e247c2.onnx,
SHA-256 10e247c2449d9a606eb5712515afda2073468b5e04e130ac898fdd70ff27136b.

The model was trained on the DIS5K dataset, whose terms of use are the dataset's own
(https://github.com/xuebinqin/DIS).