search

zhongyy / SFace

Code of TIP2021 Paper《SFace: Sigmoid-Constrained Hypersphere Loss for Robust Face Recognition》. We provide both MxNet and Pytorch versions.
86 stars 11 forks source link

Face Width and Height #5

Closed lynnwilliam closed 1 year ago

lynnwilliam commented 1 year ago

I've read the paper and the code example too. It doesn't specify the face image width and height that the input Face Mat needs to be. I have existing Faces Images and only need the Embeddings vectors for this images. So given that I have a folder with aligned and cropped faces already, what should the size of those faces be to get a good embedding vector ??

In some examples it says the Faces should be 320x320, others say 112,112 etc. What was the actual size of the faces used?

lynnwilliam commented 1 year ago

112 x 112 is the answer. its inside the ONNX model for those who are wondering.