PicMilo portrait background removal

MODNet: Real-Time Trimap-Free Portrait Matting via Objective Decomposition
Authors: Zhanghan Ke, Jiayu Sun, Kaican Li, Qiong Yan, and Rynson W.H. Lau.
Original project: https://github.com/ZHKKKe/MODNet
Original project revision: 28165a451e4610c9d77cfdf925a94610bb2810fb
License: Apache License 2.0; see MODNet-Apache-2.0.txt in this directory.
The upstream license covers code, models, and demos, excluding doc/gif assets.
No upstream GIF assets are distributed with PicMilo.

ONNX conversion and quantized weights: Xenova/modnet
https://huggingface.co/Xenova/modnet
Revision: fa2fa546052fba4c08921230a26cc69a333fca12
The model card declares Apache-2.0; a copy is included as MODNet-model-card.md.
PicMilo distributes onnx/model_quantized.onnx without modifying its bytes,
renamed to models/modnet-quantized.onnx. The preprocessing and compositing
integration are PicMilo code and do not modify the supplied model weights.

ONNX Runtime Web 1.20.1
Copyright (c) Microsoft Corporation. All rights reserved.
https://github.com/microsoft/onnxruntime
License: MIT; see onnxruntime-MIT.txt in this directory.
Dependency notices: onnxruntime-ThirdPartyNotices.txt in this directory.
The distributed runtime assets are unmodified files from the official npm
onnxruntime-web 1.20.1 package. Runtime and WebAssembly files are one build.

Asset sources, sizes, versions, and SHA-256 hashes are listed in
../vendor/background-assets.json. Model inference runs locally in the browser.