It has been a long time that neural networks learned to generate high-quality photo and even video images. My machine learning model distinguishes between an existing person from one generated by a neural network. As a human, most likely you will have a hard time doing it. Check this [https://www.whichfaceisreal.com/] sample test.
In the attached files to this task you will find data sets containing thousands of faces: some of them are real, while others are fake, that is, generated by artificial intelligence.
The fake recognition technology is very useful for security solutions, for checking media publications, for verification of user of various services and social media, and, paradoxical as it may sound, for improving the performance of neural networks that generate the images. They, in turn, have a huge potential to be used in cinema, games, social media and other spheres.
-
The data256faces.tar.gz image archive, where you will find two data sets:
a. train is the training sample, which you will use to train your model (the sample size is 8,000 images);
b. test is the testing sample, for which you will need to make a prediction (12,000 images).
-
The train.csv file for model training and parameter setup, which contains the right answer for every image from the ‘train’ (1 - fake image, 0 - real image).
dataset: https://drive.google.com/file/d/1m9hzPXQ1ffEoJEE5_Swo8WhsIHN11aP6/view?usp=sharing