FaceRecognition.Create working on memory (taking byte[] as parameter)

[ADD-eNavarro]

Hello Takeuchi-san.

I have created a face recognition system using your wonderful work, thank you very much.
Only, my enterprise works with microservices, and in order to implement the f.r. system into one I would need to work independently of the file system. There, I stumble upon FaceRecognition.Create, which only takes a folder path as input.
I've been looking into it and basically what Create does is deserialize all 4 model files using DlibDotNet. Deserialize method in DlibDotNet.Dnn.LossMmod.cs and .Dnn.LossMetric.cs both take a file, check if it exists and turn it into a byte array, then they do the proper deserializing.
I was wondering, could it be possible to create an overload for these Deserialize methods that takes directly a byte array, and extend that up to the Create method, so that FaceRecognition.Create(byte[] resnetModel, byte[] faceDetector, byte[] 5faceLandmarks, byte[] 68faceLandmarks) could be used as an overload (along with the already existing method print)?
I really love your work and would like to keep using it as you publish it, without internal tinkering.

Thank you again.

[takuya-takeuchi]

@ADD-eNavarro
Can not agree with your suggestion.
But I can add the following interface

var model = new Model {  ResNet = File.ReadAllBytes("test.dat"), ... } ;
var fr = FaceRecognititionDotNet.Create(model);

Because

• I do NOT want to add new argument if FRDN support new model file

However

• DlibDotNet does not support deserialize from byte array

So we have to update DlibDotNet at first.

Thanks.

[ADD-eNavarro]

Hi again!

Your solution seems right to me.
Are you going to open PR in DlibDotNet to implement deserialization from byte array?

[takuya-takeuchi]

@ADD-eNavarro
Yes. Just start to implement deserialize.
https://github.com/takuya-takeuchi/DlibDotNet/tree/deserialize

But I should take a long time for long long build time.
Because DlibDotNet.Native and DlibDotNet.Native.Dnn uses numerous template class.
Please be patient.

[ADD-eNavarro]

Hello again.

I will be patient, I know good work takes time.
But my boss wants me to sketch a roadmap for the deployment, so I was wondering: would it be possible to have some idea of how long are we talking about? Weeks, months? Any information will be very useful.

[takuya-takeuchi]

@ADD-eNavarro
9a3f39a

It could work on win and Linux.
osx can not pass test due to dependencies.
So I can't publidh nuget package now.

[takuya-takeuchi]

https://github.com/takuya-takeuchi/FaceRecognitionDotNet/wiki/Quickstart#1-initialize

[ADD-eNavarro]

Hello again.

With a colleague we're trying to create the NuGet package from the development branch (not using your script), but we get an error, missing DlibDotNet native dlls. So we were wondering if you could help on this matter, maybe even create the NuGet package yourself from this branch (and yes, we are aware it wouldn't work on osx).
Or maybe point us in the right direction to do it ourselves.

[takuya-takeuchi]

@ADD-eNavarro

but we get an error, missing DlibDotNet native dlls.

You can pick them from nuget file.
https://www.nuget.org/packages/DlibDotNet/

[takuya-takeuchi]

@ADD-eNavarro
Release 1.3.0.3
https://github.com/takuya-takeuchi/FaceRecognitionDotNet/releases/tag/1.3.0.3