Background-removal assets prepared for Madartools

Model: U2NETP, U²-Net lightweight general salient-object detection.
Original project: https://github.com/xuebinqin/U-2-Net
Pinned original project revision: ac7e1c817ecab7c7dff5ce6b1abba61cd213ff29
Original project license: Apache License 2.0. Full text: u2net-license.txt.

Unmodified ONNX conversion distributed by rembg:
https://github.com/danielgatis/rembg/releases/download/v0.0.0/u2netp.onnx
SHA-256: 309c8469258dda742793dce0ebea8e6dd393174f89934733ecc8b14c76f4ddd8
Bytes: 4574861
RemBG reference revision: c5b48b0417a0254df7d461efa14573597f369a83
https://github.com/danielgatis/rembg/blob/c5b48b0417a0254df7d461efa14573597f369a83/rembg/sessions/u2netp.py
https://github.com/danielgatis/rembg/blob/c5b48b0417a0254df7d461efa14573597f369a83/rembg/sessions/base.py
Reference pre/post-processing implementation: MIT. Full text: rembg-license.txt.

Runtime: ONNX Runtime Web 1.31.1, Microsoft, MIT.
Official npm package: https://www.npmjs.com/package/onnxruntime-web/v/1.31.1
License: https://github.com/microsoft/onnxruntime/blob/v1.31.1/LICENSE
Third-party notices:
https://github.com/microsoft/onnxruntime/blob/v1.31.1/ThirdPartyNotices.txt
Full local texts: ort-license.txt and ort-third-party.txt.

The runtime is configured for the WASM execution provider with numThreads=1.
No GPU, remote inference API, account, image upload or model mutation is required.
The small model is designed for general salient-object masks. It does not
guarantee precise hair, transparent materials, multiple subjects or fine holes.
Manual review/refinement is appropriate before using an exported image.

The research fixtures in this folder are upstream test photographs and are
for internal QA only. Do not publish them as product-owned images or bundle
them in the public application without reviewing their individual rights.