oesllelucena / fasnet Goto Github PK
View Code? Open in Web Editor NEWCNN archictecture for face anti-spoofing
License: MIT License
CNN archictecture for face anti-spoofing
License: MIT License
I found that the uploaded model pretrained from replay/3dmad database don't work well in other database, such as NUAA.
What's worse, I found that images with dark light are easy to be seen as real face, and images with bright light are always to be seen as attacks. Therefore, in 5105 real-face images from NUAA database, only 18% images are predicted as real face.
What can I do next? Maybe feeding more data into network? Or, just classifying one frame really works in the real world?
somebody help me? PLEASE
I'm new at this and don't have dataset for trainning,so how to get model.json file for testing?
Thank you so much!
In NUAA dataset, all test images return 1.
In all real face images, 2735 images with prob=1, and the remaining 627 images with prob>0.9.
In all attack images, 1903 images with prob=1, and the remaining 3858 images with prob>0.9.
What's a terrible performance!
I got the REPLAY_ATTACKS dataset and found that is all video, but the inputs of FASNets should be pictures,right? I transfer the video to the picture per 20 frame,and test the FASNets, it can not achieve the HTER in your paper.Can you tell me how you transfer the video to the inputs of FASNets?
@OeslleLucena
Hi, will the REPLAY-ftweights18.h5 and 3DMAD-ftweights18.h5 these two weights is already trained weight file? Can be used directly, or need to combine with VGG-16 weights and then training?
There are two documents, is tf, or th training out?
Hello, we are very interested in your open source code, I don't know if I can cooperate. We are a start-up company from China and are currently working on face recognition research and development. We are willing to pay some compensation for research and development. I hope we can cooperate. Live detection is used on Android. If you are willing to cooperate
I am working on face anti-spoofing attack recently. I have tried lots of methods, including LBP, HOG + SVM, CNN and so on. However, I found that the models trained on casia dataset and replay attack dataset are not generalized very good, cross-validation results are not satisfactory, I also try to use your per-train VGG model, also have the same problem
I just use the FASNet.py,run on Keras(Using TensorFlow backend) using the weights REPLAY-ftweights18.h5.I Pre-processing the input data ( find the face and cropped to a window sized 96 pixels ).But the result always return 1.Does anyone know why?
How do I test the code in realtime video?
The project achieved 100% accuracy of 3DMAD, and only used the color image for training. Wonder how much the depth data will help for further improvement. For example, why iPhone X's face recognition uses depth image.
Thanks
can you explain how to arrange train and validation dataset?
Layer weight shape (3, 3, 3, 64) not compatible with provided weight shape (64, 3, 3, 3)
Hello, OeslleLucena,
I'm a research student in Shanghai, China. Recently I'm interested in anti-spoofing and glad to see that your paper been accepted at ICIAR 2017. I'm really interested in your algorithm, but there still remain some questions while watching the code. So could you please let me read your paper? I'd really appreciate that if I can read it carefully.
Thank you very much.
Sincerely,
Gaussic
Sorry to bother you. The pictures I verified are all assigned to the same category. What are your suggestions for defining the parameters? Thanks for your help.
I got the REPLAY_ATTACKS dataset and found that is all video, but the inputs of FASNets should be pictures,right? I transfer the video to the picture per 20 frame,and test the FASNets, it can not achieve the HTER in your paper.Can you tell me how you transfer the video to the inputs of FASNets?
I get a error when i run test.py : ValueError: Dimension 0 in both shapes must be equal, but are 25088 and 4608. Shapes are [25088,256] and [4608,256]. for 'Assign_84' (op: 'Assign') with input shapes: [25088,256], [4608,256]
I just tried 3DMAD-ftweights18.h5... always returns 1 on 96x96 centered face image (as @OeslleLucena recomended)
Just tried REPLAY-ftweights18.h5... always returns 0. Same pair of original / spoofd image.
I am trying to open REPLAY.h5... model
model = keras.load_model(path)
and it throws
ValueError: Cannot create group in read-only mode.
Hi Oeslle,
Great work with FASNet! However, I tried running the model with some test images, and they always seem to be detected as fake. After reading your paper carefully, I found this:
In step two(face detection), we first used the OpenFace face detector [35] algorithm to find the region-of-interest (ROI) corresponding to a face. Next, using OpenFace algorithms we cropped to a window sized 96 pixels and aligned the faces to center based on the nose and eyes position.
How can I replicate this procedure for my own test images? Can you possibly share the code used to detect the face, crop the image and center align the images based on nose and eyes position?
Thanks for sharing your code.
I'll train the model myself with tensorflow.
How can I get the train image dataset?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.