Deface.NET 10.0.2

Deface.NET: Video and photo anonymization library for .NET

Deface.NET is an MIT licensed library for photo and video processing to achieve anonymized resources. It is based on Ultraface ONNX model for face detection, custom trained YOLO-NAS-S ONNX model for license plate detection and SkiaSharp + FFMpeg image processing.

Installation

Deface.NET is available via NuGet package:

nuget install Deface.NET

Also, you need to install FFMpeg and FFProbe manually.

Using Deface.NET

With Dependency Injection (recommended):

Deface.NET can be easily added via Dependency Injection:

services.AddDeface(settings =>
{
    ...
});

and then IDefaceService can be easily injected:

public class SomeDIClass
{
    private readonly IDefaceService _defaceService;

    public SomeDIClass(IDefaceService defaceService)
    {
        _defaceService = defaceService;
    }
}

With Standalone Provider

Another way of using Deface.NET is to use DefaceProvider:

IDefaceService defaceService = DefaceProvider.GetDefaceService(options =>
{
    ...
});

var result = defaceService.ProcessImage("1.png", "1-processed.png");

GPU Compatibility

• You need to have cuDNN and CUDA Toolkit installed.
• See NVIDIA docs for CUDA support matrix.
• See tested NVIDIA configurations for Deface.NET.

Usage

services.AddDeface(settings =>
{
    ...
    settings.Hardware = Hardware.Cuda(gpuId); // for most configurations gpuId = 0 or gpuId = 1 works
    ...
});

Configuration options

To find out more informatiom, read the XML comments for a given option property/value.

Commercial Features

Deface.NET's default settings run the library in open-source mode. However, the library includes some features that are licensed differently. Ensure you have the proper license before using them. Deface.NET's creators take no responsibility for the use of commercial features without the necessary rights.

Each enabled commercial feature displays information about its usage, including additional links to the license, etc. You can hide this information by setting HideCommercialFeaturesInfo = false in the settings.

The current list of commercial features requiring an additional license:

FAQ & Known Issues

1. Long paths on Windows

Videos may not be processed correctly on Windows if the destination path exceeds 260 characters. See Microsoft Docs to learn how to enable long paths.

2. GPU Performance

Currently, the Ultraface model used in Deface.NET achieves only ~30% performance improvement when running on a GPU. We are working on optimizing it. NOTE: GPU is strongly recommended with license plate blurring - current model is poorly optimized for CPU usage.

3. CustomBaseDirectory

In some cases (such as with Azure Functions), Deface.NET may throw a FileNotFoundException indicating that Settings.CustomBaseDirectory should be overridden. This happens because the native static method AppContext.BaseDirectory from the System namespace may not return the actual directory containing the application DLLs.

To resolve this, set the CustomBaseDirectory property when registering Deface.NET. In such cases, applications often provide the correct application directory as an environment variable.

4. Missing ONNX Files

In case your application throws an exception about missing .onnx files (most like due to Docker container), to your .csproj file add:

<Target Name="CopyResourcesToPublish" AfterTargets="Publish">
    <ItemGroup>
        <GeneratedResources Include="$(OutputPath)Resources\**" />
    </ItemGroup>

    <Copy SourceFiles="@(GeneratedResources)" DestinationFolder="$(PublishDir)Resources\%(RecursiveDir)" />
</Target>

5. Docker Support

nvidia/cuda:12.4.1-cudnn-runtime-ubuntu22.04 image is tested inside production environment and works correctly.