Help! Can I build my FYP using this ? #138
-
I want to use human to build a website where users enter details of other people along with their pics and Guest Users can upload an image to find similar people by their faces. as new entries and pics will be added on the go, Will face models need to be trained on every added pic ? How would be performance ? How you would recommend implementing, as this is my first project ? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 2 replies
-
There is no "training", you only need to run detection on each uploaded photo and store resulting face descriptor (which is an array of float numbers). Assuming using NodeJS for backend implementation, backend would preload all required models and wait for request from frontend. Then you store the image and corresponding face descriptor somewhere on the server (as a file or in a DB) and also keep that face descriptor in an array (each new descriptor gets added to array). Final step would to loop through the array and run face match between current descriptor and each descriptor in the array, sort results them by similarity and send top-n results back from server to frontend. Performance-wise, running detection and extracting face descriptor is measured in miliseconds, so thats not a problem. Now, to get full performance from the backend that operation should be performed in a separate worker as NodeJS is by definition single-threaded - you don't want one user to hog entire server. You can check included demos for an example of NodeJS multi-process implementation. |
Beta Was this translation helpful? Give feedback.
There is no "training", you only need to run detection on each uploaded photo and store resulting face descriptor (which is an array of float numbers).
Assuming using NodeJS for backend implementation, backend would preload all required models and wait for request from frontend.
As user uploads the picture to frontend, it sends it to server which just runs face detection and extracts face descriptor.
Then you store the image and corresponding face descriptor somewhere on the server (as a file or in a DB) and also keep that face descriptor in an array (each new descriptor gets added to array).
Final step would to loop through the array and run face match between current descriptor and each…