Face Anti-Spoofing for Facial Recognition in Python

[@tristdrum-workweek]

Thank you Sefik! You're contribution to the world is wonderful. I appreciate what you are doing greatly!

[@masterspirit]

Serengil, we love you! God bless you!

[@MicheleSblendorio]

Thank you! A really useful and interesting implementation of DeepFace. Ad maiora semper!

[@ramyzeidan]

Thank you for your video, but im keep getting spoofing even though im using my webcam on me no phones or paper detected? is there away to optimize it to get accurate results? face_objs = DeepFace.extract_faces(img_path=face_img, anti_spoofing=True) for face_obj in face_objs: if not face_obj["is_real"]: # Spoofing detected, draw red rectangle cv2.rectangle(image, (x, y), (x + w, y + h), (0, 0, 255), 2) response_data["message"] = "Spoofing detected." return response_data

[@sefiks]

That is the most robust model i found :/

[@Shaurya-dz4rd]

Hey, can you please tell which loss function is used to train the VGGFace and Facenet models. I require them to build the fine tune pipeline. Thanks!

[@sefiks]

These are pre-trained models, I didn't train them. You have to investigate this from reference repos which are mentioned in the read me of the deepface.

[@JohnCristophereCarlos]

Are there a way to translate this library in javascript or web version, I have a web app and mobile app based in webview and wanted to implement this kind of liveness. Is it possible?

[@sefiks]

Deep Face Recognition With ReactJS and DeepFace: kzbin.info/www/bejne/f4nSkptsp82rl8U

[@JohnCristophereCarlos]

@@sefiks Thank you for the tutorial, this is great. My next question is that given that this is based on the API you call in the backend. On top of my mind is an approach wherein all the valdation for anti-spoofing is within the devices no need to hit the API endpoint so that it can support the offline mode.

[@f1ras217]

can anti spoofing be used with the api?

[@sefiks]

Yes, it can.

[@f1ras217]

@@sefiks what should i do? i already set the anti_spoofing = true but it still return the verification result as true. it should be false right if i put the spoofed image

[@sefiks]

@@f1ras217 it will throw exception if it is fake

[@f1ras217]

@@sefiks okay last. what kind of fake image that the model will throw exception? i'm currently using face verification. i tried one image is real, captured using the camera and another one is an image that has been captured from another device and i uploaded it into deepface. is that still count?

[@sefiks]

Try the one i tested in the video

[@f1ras217]

brother, the exception is not thrown inside the api. if i compare image 1 (fake) and image 2 (real), it will not throw any exception and proceed with normal verification like usual. when i tested to compare verify from ipynb, it threw the expected exception. i wanted to use the api because i'm developing a system that utilizes deepface. thank you in advance.

[@sefiks]

are you using v92? when i tested it as shown, I am getting the following response. I will add more details to the response message tonight. endpoint: localhost:5005/verify body: ``` { "img1_path": "/Users/sefik/Desktop/deepface/tests/dataset/img1.jpg", "img2_path": "/Users/sefik/Desktop/f1.jpeg", "anti_spoofing": true } ``` response: ``` { "error": "Exception while verifying: Exception while processing img2_path" } ```

[@f1ras217]

@@sefiks yes, i'm using v92 24-06-11 16:58:47 - Welcome to DeepFace API v0.0.92! endpoint: 127.0.0.1:5000/verify body: { "img1_path": "C:/Users/fiira/face_verification/elul.jpg", "img2_path": "C:/Users/fiira/face_verification/spoofing_example.jpg", "anti_spoofing": "True" } response: { "model": "Facenet", "similarity_metric": "euclidean", "threshold": 10, "time": 32.99, "verified": true } it is still detecting like usual... i'm still figuring out, where did i go wrong 😭

[@f1ras217]

@@sefiksyes, i am currently using v92. when i launched the api: 24-06-11 16:58:47 - Welcome to DeepFace API v0.0.92! endpoint: localhost:5000/verify body: ``` { "img1_path": "C:/Users/fiira/face_verification/elul.jpg", "img2_path": "C:/Users/fiira/face_verification/spoofing_example.jpg", "anti_spoofing": "True" } ``` response: ``` { "model": "Facenet", "similarity_metric": "euclidean", "threshold": 10, "time": 6.65, "verified": true } ``` i'm still figuring out, where did i go wrong.

[@f1ras217]

@@sefiks yes, i am using v92. i used the same header as you did, with my sample spoofing images. but the response is like this "model": "VGG-Face", "similarity_metric": "cosine", "threshold": 0.68, "time": 5.23, "verified": true

[@f1ras217]

ohhhh so basically, if there is a spoofed image has been inserted, the api will return 400 code response?

[@sefiks]

Yes

[@yuryeuceda8590]

@@sefiks I did a change locally that returns a property with false instead of an exception because in my case is more useful. I have a really really urgent question, can you readme in the comments please

[@yuryeuceda8590]

Hello, great job Sefik, Just now I need to make a fake face detection but when using deepface in my computer it works great, but I need to do it on a serverless function and first, the size of image get big (2GB ~ 8GB) and the memory requirements too (8GB). I want to know how to remove all code except antispoofing module because thats what I only need. Thanks a lot. Currently, we have a serverless function using dlib that performs face detection and face embedding but we need to complement it with another serverless function to make fake face detection score and joining this, we create an integrated service, but Deepface seems to be too big in order to be deployed as a serverless function.

[@sefiks]

@@yuryeuceda8590 then deepface may not be the right tool for you. Recommend to use anti spoofing repo directly: github.com/minivision-ai/Silent-Face-Anti-Spoofing

[@yuryeuceda8590]

@@sefiks Thanks friend for your quick answer

[@humamalfariq1702]

Great i like it

[@TechInstructorHub]

i am not able to get green box

[@sefiks]

@@TechInstructorHub getting red or gray?

[@TechInstructorHub]

@@sefiks just web is open but nothing display like no red and green box

[@sefiks]

Seems your installation is problematic

[@TechInstructorHub]

@@sefiks I have cross check my installation and also update the deepface but not working properly

[@TechInstructorHub]

@@sefiks i am create a project. Face Attendance, in this project can i use the deepface in this project with the django framework. if i am performing i need to take deepface all source files code?

[@tunateke9806]

Cool