Build a Deep Facial Recognition App // Part 7 - Real Time Predictions with OpenCV // #Python

Ever wanted to implement facial recognition or verification into your application?

In this series you'll learn how to build a deep facial recognition application to authenticate into an application. You'll start off by building a model using Deep Learning with Tensorflow which replicates what is shown in the paper titled Siamese Neural Networks for One-shot Image Recognition. Once that's all trained you'll be able to integrate it into a Kivy app and actually authenticate!

In Part 7 you'll go through how to:
Setup Verification Images
Build Verify Function
Perform Recognition in Real Time using OpenCV

Links

Chapters:
0:00 - Start
0:29 - Explainer
1:09 - Tutorial Start
1:48 - Whiteboard
5:09 - Setup Verification Images Folder
7:53 - Build Verification Function
15:30 - Make Predictions
17:08 - Calculate Detection and Verification Thresholds
20:32 - Access Webcam
26:30 - Add Verification to Loop
31:32 - Testing the Final Model

Oh, and don't forget to connect with me!

Happy coding!
Nick

P.s. Let me know how you go and drop a comment if you need a hand!

[mikecooper]

I love ur motivation Nick! Another great video.

[nhlakaniphomagwaza]

Thank you Nicholas... Looking forward to the final episode !!!

[akshitdayal]

Awesome series! Thanks for all the effort put in...

[sebastiansulz]

Will you be making a spoof face detector model in the next part/series, so that the model will not authenticate if someone tries to gain access through our photograph or video?

[microgamawave]

What about 3D face recognition?? And how to convert two cameras in to stereo camera?? For creating 3d face objects And Distance measurement which

[ashketchum]

Thanks for the series, I learned alot. Would be great to have a section about hints / suggestions to extend it further in the next video.
E.g. What we need to change to register multiple users

[TAHMEED_REZA]

I seem to be having issue getting a match as soon as the background changes. I think if you segmented out the faces from the images (anchor, positive and negative) maybe this would work even better.

[nur-e-jannatanika]

I did the exact same things you did but my model always give True output, even if I'm giving a Google picture as an input. Can you please tell me why it' snot working for me and how to make it work? Also, wonderful tutorial- your explanations are so well-put-together!

[hannav]

can't wait for the final part : > do you think it's possible to put all parts into one long video?

[mtalhakhalid]

I use an Arduino library and then use my trained dataset so if my ace verified it blink led in the future use it to open the door all happen thanks to you sir

[srishtichaubey]

I want to do one shot learning. Keeping 50 images of each user is costly. I just want to keep one image, and match with it.

[goboy]

I massaged the "lfw" data set to get 480, 000 labelled image pairs. It would be nice if I could do training in parallel on multiple cores. Any suggestions on how to make that happen?

[howtodoitdifferently]

I did perform it and it does work when tested with my present images.. but when I tested it on old images it does not work.
So, do i need to feed my old images as well..
Isn't this a issue.. what can i do here in this case. feed my old images?? any other suggestion

[balajiv]

Please create a federated learning project 😋..look in to it for us, it would be a great help mate there is not much content available on this topic other then from tenserflow network itself🙌

[dGDeika]

i thought face recognition model works only with input only and tries to find certain person with all faces that the model has been processing. Maybe there's a different approach to this not only matching two images. Now, at least, i understand how face id verification could work in theory.

[finn-gs]

Thanks for the videos. I wanted to as whether this a triplet network or a siamese network?

[AkshatDarji]

Hi, your tutorial for the Siamese network is very good. I have to use this saved Siamese network model for the person Identification from CCTV footage so which kind of changes I required to implement it. brief of this is that I have to identify the student from the CCTV footage and show the label of the detected student i.e. roll number of that student.

[singh.srishti]

Hello Nicholas,
I completed this whole series until part 7, and everything worked fine, just that when i run the inference real time, it detects the negative images as True??
What might be problem?
Since there’s no error I am unable to debug, my guess is it might be because of input images which has been taken from a century old webcam.
Please help?

[Meego_]

how do i make it for multiple persons ? do i have to train network everytime i add a new person ?

[rohitchan]

Hey Nicolas, can you please make a series on MDPs(value iteration/policy iteration), model free learning, Monte Carlo tree search, Q learning?. I'm doing my master's in AI so it'll be very helpful for students like me.