Any chance of a arm version nuget package? about facerecognitiondotnet HOT 7 OPEN

It's not FRDN issue but DlibDotNet. But I can understand what you hope.
There is 2 problems.

• how to build dlib by using arm
• it could be helpful https://stackoverflow.com/questions/45158511/how-to-enable-neon-acceleration-with-dlib-on-arm
• can i support arm package for nuget
  • it could be helpful https://docs.microsoft.com/ja-jp/dotnet/core/linux-prerequisites?tabs=netcore2x

from facerecognitiondotnet.

Should I have posted in the DlibDotNet issues then instead? I understand that the wrapper for dlib and this FaceRecognitionDotNet are targeting .net standard so that's not an issue.

If it's helpful I've already done a lot of reading on the subject and you may find some of the resources helpful.

• How to build dlib by using arm?

  • Cross compiling dlib for ARM on Ubuntu https://www.jofre.de/?p=1494
    • When I wran the build command in that article I ran into this error that I wasn't able to resolve: Parse error in command line argument: -DENABLE_NEON” I found a github issue where someone seems to have had the same error and resolved it but didn't mention how. I tagged them with a question about it: davisking/dlib#549 Maybe you would be able to shed some light on this.
  • Article about compiling dlib on an actual Raspberry Pi apparently works after increasing the swap size: https://www.pyimagesearch.com/2017/05/01/install-dlib-raspberry-pi/

• Enabling neon when compiling for arm is apparently functional on the Raspberry Pi 3 based on these comments: davisking/dlib#276 (comment)

• Nuget support for arm

  • I haven't personally compiled my own nuget packages with c++ dependencies before but this other project has done it with a package supporting arm so I believe it's probably possible: https://www.nuget.org/packages/TensorFlowSharp.ARM
    • Here's the merge for that project where you can see specifications for a .targets and the .csproj for the arm version of the nuget package: migueldeicaza/TensorFlowSharp@aa2c339

from facerecognitiondotnet.

I was success to build DlibDotNet by arm64 on ubuntu 16.04.

# ldd build_linux_arm64/libDlibDotNetNative.so
        not a dynamic executable
# file build_linux_arm64/libDlibDotNetNative.so
build_linux_arm64/libDlibDotNetNative.so: ELF 64-bit LSB shared object, ARM aarch64, version 1 (GNU/Linux), dynamically linked, BuildID[sha1]=7dd15217150b68eede19a8d3c03216bedee7523c, not stripped
# ldd build_linux_arm64/libDlibDotNetNativeDnn.so
        not a dynamic executable
# file build_linux_arm64/libDlibDotNetNativeDnn.so
build_linux_arm64/libDlibDotNetNativeDnn.so: ELF 64-bit LSB shared object, ARM aarch64, version 1 (GNU/Linux), dynamically linked, BuildID[sha1]=e58b054fe3e5e781ce66ad00868a31446a6e3c23, not stripped

But arm64 is not supported on dotnet core 2.0 and arm is supported but it may be able to work with high performance.
dotnet core 3.0 is preview release for now.

However, I created DlibDotNet-ARM nuget package.
Perhaps, I will be able to create and test FaceRecognitionDotNet-ARM on raspberry pi 3 after install dotnet core 3.

from facerecognitiondotnet.

That's fantastic!

Also interesting, I was unaware of the lack of arm64 support by dotnet core however I see this issue outlining the progress for supporting it in the 3.0 preview like you mentioned:
dotnet/announcements#82

Anyways though I'm super excited to try it out your nuget package! I looked on nuget and didn't find it at https://www.nuget.org/packages?q=DlibDotNet-ARM
Have you uploaded it yet?

And more than anything though I'm still most excited about the prospective FaceRecognitionDotNet-ARM package. Are there any particular barriers that stand in the way of that effort or is it just a matter of testing on a raspberry pi?

from facerecognitiondotnet.

I have raspberry pi 3 installed raspbian.
but i found that official raspbian force device to run on 32bit mode even though it have Cortex-A53.
I haven't told you this yet, I built 32bit arm binaty.
However, I doubt 32bit binary is high efficient.

So I didn't test nuget package I created.

from facerecognitiondotnet.

I published FaceRecognition ARM.
It is alpha release. It did not test yet.
But it may be useful.
https://www.nuget.org/packages/FaceRecognitionDotNet-ARM/

from facerecognitiondotnet.

I published FaceRecognition ARM.
It is alpha release. It did not test yet.
But it may be useful.
https://www.nuget.org/packages/FaceRecognitionDotNet-ARM/

Today i had test your great ARM version, it was OK with little code change. happy to know that we can run on 'Rockchip 3328/3399' two boards, (ubuntu 16.04), either face detect or landmark computing is OK. the only problem is ... SLOW :) (may be we can speed up this by using multi CPU core in the future)

from facerecognitiondotnet.