Face recognition iOS+Swift+OpenCV (or any other framework) - measure face similarity

Hello, Here I’ll describe the task in more details so you can have an idea about the scale of the project, requirements, flow, and can give an approximation of the time and price. The app will receive all the pictures from Facebook profiles, that means that they can contain a lot of different poses, angles, lighting, can have several people on a picture, upside down picture – but clearly most of all will be more or less normal and representative, but nevertheless, I’d like to capture as many possible pictures as possible, including all the ‘difficult to recognize’ pictures. So, next steps: pictures of different users will be shown to a primary user like in a dating app, so the user can either Like or Dislike another user. Thus, we will have two groups of profiles, people who have been liked and people who have been disliked. After that, as soon as we have enough of images in both group (we should also determine what ‘enough’ means), starting from N-th profile, the profile can be compared to the 2 groups; I have two options here, first is – measuring average similarity between a picture and a group, so if average similarity of the Like group is bigger than in the Dislike group – then a profile of that user must be shown to our primary user; second – the same flow but measuring the maximum similarity in a group, rather than mean – so if the maximum similarity with a single person in the Like group is bigger than in the Dislike group then show… You got the idea. The gist of the app is to show to the user only those pictures and profiles that have a big probability of being liked; depending on the time and price there are at least 2 additional ideas using machine learning and one other feature. Answering some possible questions: there is a ready app template written in Swift that works; first the face has to be detected on a picture, normalized, and having all possible manipulations to boost the accuracy of the recognition part, and only after that recognized; the app will use only pictures, no video; the time to process one picture should be not more than one second – including detection and recognition part, but the less the better. The database of pictures will be pulled out from Facebook. The database for comparison with a target face can be dynamic or cumulative, so it can be updated after every new profile in a group with excluding some previous profiles or accumulating profiles up to some number. First of all this should be an iOS app, but I also have app templates for Android as well, they both work and ready to use, they just lack the face recognition part, also some changes in the code will be needed to integrate recognition, the use of API, deletion of some code blocks related to creation a server as I won’t need one, and some minor design changes. But nevertheless, the main focus of the app should be concentrated around the recognition part; the user should be provided with the best experience possible, so the more accurate algorithm is, the more reliable, robust the system is, the better it is reflected in user’s experience of the app. Considering all the information above, please reply in details about the algorithms you will use for this project, how accurate they will be considering all the possible profile pictures, what percent of faces in such a random database of faces could be detected and recognized using your system, what are some critical values of the angle/pose/rotation/etc. that can be allowed – at least a rough estimation. Can you improve your current algorithms to increase accuracy? Can you use different algorithms with better accuracy? Finally, your time and price approximation (it doesn’t have to be exact number, just some abstract level for me to compare quotes; it can also be hourly rate with estimation of hours or fixed price for the project). I won’t consider candidates that haven’t answered all the questions above, these questions are crucial for the project.