Giant Tiny Background — third-party notices

U²-NetP model
Original authors: Xuebin Qin, Zichen Zhang, Chenyang Huang, Masood Dehghan,
Osmar R. Zaiane and Martin Jagersand.
Project: https://github.com/xuebinqin/U-2-Net
Paper: U²-Net: Going Deeper with Nested U-Structure for Salient Object Detection,
Pattern Recognition 2020. https://arxiv.org/abs/2005.09007
License: Apache License 2.0; full text in U2NET-APACHE-2.0.txt.
The unmodified u2netp.onnx graph/weights are distributed by Daniel Gatis's rembg:
https://github.com/danielgatis/rembg/releases/download/v0.0.0/u2netp.onnx
Bytes: 4574861
SHA-256: 309c8469258dda742793dce0ebea8e6dd393174f89934733ecc8b14c76f4ddd8
This is the small general salient-object model, not the separately trained
portrait or human-segmentation variants. The host does not fine-tune or change
these weights. Model assets are served from this site and run on the user's device.

ONNX Runtime Web 1.22.0
Copyright (c) Microsoft Corporation. All rights reserved.
License: MIT; full text in ONNXRUNTIME-MIT.txt.
Additional upstream notices: ONNXRUNTIME-NOTICES.txt.
Source: https://github.com/microsoft/onnxruntime/tree/v1.22.0
Package: https://registry.npmjs.org/onnxruntime-web/-/onnxruntime-web-1.22.0.tgz
Only the unmodified WASM-only JS module, WASM loader and WASM binary are bundled.
No GPU/JSEP backend, proxy worker, optional provider, remote CDN, analytics,
service worker, or persisted user-image cache is used.
The central release builder may append release queries to module references;
this changes cache identity only, not the vendor runtime logic.

Preprocessing reference
The U²-NetP input normalization and output min/max normalization follow the
public rembg session contract. rembg is MIT-licensed; REMBG-MIT.txt retained.
https://github.com/danielgatis/rembg/blob/main/rembg/sessions/u2netp.py
https://github.com/danielgatis/rembg/blob/main/rembg/sessions/base.py
This application implements its own browser adapter. It does not include rembg.

Exact installed-asset SHA-256 hashes and package integrity are recorded in
../provenance.json in the source and release archive.