Original: the extension of free and open Christie masks face detection and recognition technology
Thinking
Wearing masks incorporated by 20% or more specialized recognition training dataset recognition model wearing a mask, and then adjusting the overall recognition Pipeline detect wear masks for the first, if it is determined as a recognition model wearing a mask with a mask is used for identification, otherwise, use a conventional recognition model for identification. In the example below:
Therefore, the entire process:
- retinaface detection + classification + key points
- The selected recognition model classification result (wear masks / not wearing masks)
SDK deployment
TenniS should Tencent ncnn, Ali mnn like.