Insightface: Train on My Own Data Set

[ahkarami]

Dear @yingfeng, @nttstar, @yuzhichang,
I have a face data set with the following format:

FaceDataSet/
    ID1/
        face_imageID1_1.jpg
        face_imageID1_2.jpg
        face_imageID1_3.jpg
    ID2/
        face_imageID2_1.jpg
        face_imageID2_2.jpg
...

Would you please kindly guide me to how I can train your models (e.g., with ArcFace Loss) on my own data set?

[hustzeyu]

You need to pack all your jpgs into rec and mxnet has that function

[eliaY]

@hustzeyu i am still a beginner in this stuff can you please provide me with the steps.
thanks

[borisgribkov]

Look at the first post here https://github.com/deepinsight/insightface/issues/256, TrillionParts data has the same format as yours. But I guess you have no landmarks for your data, so maybe some modifications of face2rec2.py will be needed.

[Ffmydy]

@ahkarami
Is your own face data set under the camera?Could you please share me?