Implement BiRefNet ONNX inference for .NET

[Copilot]

Complete ONNX Runtime implementation for BiRefNet image segmentation and background removal in .NET.

Core Implementation

• BiRefNetModel: ONNX inference with sync/async APIs, mask application, proper resource disposal
• ImagePreprocessor: Resize to 1024x1024, ImageNet normalization (RGB → NCHW tensor)
• ImagePostprocessor: Sigmoid activation, configurable threshold/smoothing, handles non-square tensors
• BiRefNetHelper: High-level utilities for background removal and mask extraction

Usage

using BiRefNetSharp;

// Load model once, reuse for multiple images
using var model = new BiRefNetModel("model.onnx");

// Simple background removal
BiRefNetHelper.RemoveBackground(model, "input.jpg", "output.png");

// Or manual control
using var image = Image.Load<Rgb24>("input.jpg");
using var mask = model.Predict(image);
using var smoothed = ImagePostprocessor.SmoothMask(mask, sigma: 2.0f);
using var result = BiRefNetModel.ApplyMask(image, smoothed);
result.SaveAsPng("output.png");

Package & Tooling

• NuGet package metadata with XML documentation
• Cross-platform CI/CD (Ubuntu, Windows, macOS)
• Sample console application demonstrating usage
• Comprehensive documentation (README, EXAMPLES, CONTRIBUTING)

Dependencies

• Microsoft.ML.OnnxRuntime 1.19.2
• SixLabors.ImageSharp 3.1.12 (latest, no CVEs)

Original prompt

BiRefNet's ONNX inference implementation under .NET

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.