How does detectAllFaces function will handle fake image photo by mobile

[danies8]

Hi, I used this code that detect all faces, I want to avoid that user will photo himself by cellular and show the cellular photo to camera , how I avoid by code ? For know fake image by cellular also work. I saw this link in history issue and there is no option do install through npm ? https://github.com/vladmandic/anti-spoofing

 const ssdOptions = { minConfidence: 0.1, maxResults: 10 };
      const optionsSSDMobileNet = new faceapi.SsdMobilenetv1Options(ssdOptions);
      const faces = await faceapi
        .detectAllFaces(tensor, optionsSSDMobileNet)
        .withFaceLandmarks()
        .withFaceExpressions()
        .withFaceDescriptors()
        .withAgeAndGender();

[vladmandic]

faceapi does not perform any analysis on the detected face - real or fake or doctored or anything
and given that faceapi here is strictly in maintenance mode, this request is out of scope

[danies8]

Thanks a lot:) I tested this link and is not solved the problem: https://github.com/vladmandic/anti-spoofing

[vladmandic]

that is just a test - a very simple model and trained on small dataset to detect real people vs fake images. it has nothing to do with photo on the phone.

[danies8]

Hi, Do you know any other lib that will solved this issue, that I can add to my solution for detecting fake images ?

[vladmandic]

i dont, sorry...

[danies8]

thanks a lot

[vladmandic]

update: i just ran into https://github.com/Prem95/face-liveness-detector
its a keras model with a lot of dead code, but could be converted to tfjs and quantized

[danies8]

Hi, My project is node js and this project is pyton , how they can come together ? Thanks,

[vladmandic]

Model is not in Python, model is in Keras. It's just that implementation is in Python

If the model converts cleanly from Keras to TFJS (there is one custom op SparseCategoricalFocalLoss that may need reimplementing, but maybe it can be skipped), the rest is easy - just ignore Python code

1. Detect face using FaceAPI
2. Crop the face from original input using tf.image.cropAndResize
3. Optionally normalize it using something like tf.div(input, 255) - all depends what value range model expects
4. Run model using model.execute and pass it cropped/resized/normalized input
5. Get results, its a float32 value that represents percentage change

[danies8]

Hi, Can you give an example how to do it ? It also good for picture not only phone ?

[vladmandic]

Can you give an example how to do it ?

Example would be to write it all. I may do it, but its not a priority.

It also good for picture not only phone ?

Looks that way.

[danies8]

Thanks a lot, we are waiting.

[danies8]

Hi, Did you have a chance to write it ?

[vladmandic]

no and not likely soon.

[danies8]

Thank you.