In relation to the COVID-19 pandemic, I'm interested if it is possible to recognize the faces of people who are wearing masks? It seems to be important now because many crimes can be committed by people in masks. Besides this, there are many other use cases for this technology.
The problem is not new. It belongs to the group of tasks called face recognition with occlusion. I have heard that it is possible and even was successfully tested and used in China.
Actually, the existing models showed not very good results (around 50% of accuracy). But after the special training, it is reported that the accuracy of recognizing people in masks has risen up to 95%. Here is an interesting material - https://www.ft.com/content/42415608-340c-4c0a-8c93-f22cdd4cc2d6
As in elsewhere in machine learning, the solution of this problem is more dependent on the size and the quality of the available dataset, than on the coding and model building. Give me the large labelled dataset of faces in masks and I will train the good facial recognition model.
I agree with @Targo_ that the dataset is critical, but it is also possible to spoil the results by doing some silly things when building the model, even if you have the huge and diverse dataset.
It seems that it is very hard for individual researchers to build such a model, given that the special dataset is required. It is the task for AI firms and startups, I think.
Yes @rEvee750 , you're right. An individual will collect needed dataset for ages. But I think given the hype around technologies that can help to fight coronavirus, the datasets with people in masks can appear in open source soon.
Thanks for sharing your thoughts!
Just drop us an email to ...
Show more 
Log in with Google