Build a Deep Facial Recognition App // Part 2 Collecting Data // Deep Learning Project Tutorial

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 2 you'll go through:
1. Collecting Negative images from Labelled Faces in the wild
2. Resizing OpenCV Output Frames for Image Collection
3. Collecting Positive and Anchor Images

Links

Chapters:
0:00 - Start
0:28 - What's Covered
1:45 - Whiteboard Session
7:34 - Collect LFW Data
12:20 - Moving Images
19:38 - Access Webcam with OpenCV
27:14 - Changing OpenCV Frame Size
32:43 - Saving Images
43:16 - Wrap Up

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!

[amritbhattarai]

Hi Nicholas, I really really appreciated your in-depth explanation line-by-line. As a beginner in Computer-vision trying to learn upon the concepts and implementations, your video tutorials have been invaluable. Thank you so much!

[edrianabadiano]

damn, why are you so underrated
pls upload more about this topic as soon as possible

[TejrajParab]

Thanks, I can't wait for next part!

[sharankalyan]

Great one. So much to learn from! Really appreciate the effort and work 👍🏻😊

[anishaudayakumar]

Awesome!! Waiting for the next Part :)

[tinashechinyati]

Thanks Nicholas. I love your videos and I can easily follow. This is my number 1 and favorite Data Science YouTube channel. Thanks a lot. Would love to see more of research paper to code videos.

[rudyitiel]

Hello Nicholas its a very interestint video an project I really loved it but I have problemes in getting the dataset how can I access it please

[goboy]

I am taking a slightly different approach here. I want to build a robot that can remember faces of strangers and not require retraining. I don't want to use any one person's face for training. So here is my approach:
The lfw collection has thousands of directories with either a single image or multiple images. - "singles" and "multiples" for short. All of the negatives come from the singles. All of the "training anchors" come from the multiples
All of the "labelled positives" come from combinations of multiples. For example, there are eight images of Tom Cruise which produces 28 labelled positives.
All of the "labelled negatives" come from matching a training anchor with a negative.

[BlakeTanyileke]

Your videos are so explicite, i like that. Thank you 👏👏

[charishmapurama]

Hey I cant find the labelled faces website what should I do?
Can I use a dataset from Kaggle instead?

[joeldjoy]

I'm not able to download the image database somehow. Whenever I click, it just loads the page and doesn't download anything. Is there any other way to get the dataset

[onndwelatshisikule]

Hi Nic, when I try to collect the images via webcam, it doesn't open and it gives no error, what should I do?

[manasisingh]

Hey Nicholas!!
While using your code to create the directories, I'm getting the error saying that they already exist, but I can't seem to find their paths in my system. Also, due to this there are many subsequent errors lining up.

Kindly help asap!! Thankyou :')

[sushilkhadka]

Why do we need negative image?
We first compare anchor with positive, If it outputs one then we can say Verified but If it'outputs zero (close to zero) we can say Unverisifed. What is the need for negative?Please help me understand this

[mohammedabed]

Thanks for this great tutorial. Lets assume that I have already defined multiple persons that I would like to include them within the dataset in positives and anchors. How can I achieve this ?

In the tutorial you were including only yourself in the positives and anchors, but what if I wanna include the faces of multiple persons (each in a single image) ? Can I just collect there images, apply data augmentation, and move them manually to the positives and anchors ?

[mtalhakhalid]

Impressive video sir i am your fan now subscribed to you thanks for such wonderful work.

[francisoyediran]

Hi, great work, however I encountered an error while trying to follow the tutorial,
for directory in os.listdir('lfw'):
for file in os.listdir('lfw', directory)

Error: listdir() takes at most 1 arguement (2 given)

[Nerfzisback]

hey nicholas, isn't better to create data of only faces ? that means collecting faces with a face detector from images

[junkimedm]

Hello admin. How to AI python convert video to jpg to text 😞

[johanvijayan]

how did you put the type as "Jupyter Source File" in file explorer before extracting the lfw file, mine comes as ipynb file type