Using a variation of One-Shot technique, it recognizes the person with just one photo.
Using more photos it will only improve the result, but try to vary the poses
Be sure to have these modules instaled, you can install them with pip install <module>
- cv2 (OpenCV)
- dlib
- imutils
- numpy
- argparse
- pickle
- face_recognition
There's a precompiled dataset on the repo, it have more than 100 celebrities registered, you can run the script directly with one_shot.py
and see the algorithm find the celebrity most like you or try it in a movie trailer(check input folder for a surprise 😉) with one_shot_video.py
(run <script-name> -h
for how-to-use)
If you want to run in a new dataset, do this:
- Put your photos in a folder in /datasets
- Run
align_faces.py
in /tools passing the path of the photos you just added (this step is for better results) - Run
generate_embeddings.py
to generate the embeddings (obviously) - Now is the time to play! 😄, run
one_shot.py
for real-time recognition using your webcam orone_shot_video.py
for recognize persons in a video file (only supports mp4 videos)
There is a version who send the proccesing results to an app on your smartphone, the application is under development and a release will be available soon!
You can track the status on this repository:
All scripts have a -h
option that will explain what you need to pass in arguments.
And you can contact me btw!
-
Our teacher for the counseling and patience
-
Every people that in some way helped us
-
These projects and software