Minyoung Jeong I went to a conference the other day, and this research was part of the keynote. Its about generating food images with GANS to train a food recognition classifier:

Preprint The Art of Food: Meal Image Synthesis from Ingredients

These are some related slides, from the same research team:

https://www.datacouncil.ai/hubfs/Data%20Council/slides/bcn19/01%20-%20data%20council%20Uncertainty-aware%20Food%20recognition%202019%20Petia%20Radeva.pdf

Hope that helps to get you started!

Lena

According to some studies, the answer is yes. GAN can improve recognition.

Here is a sample paper that uses GAN for arithmetic data generation. It's for plant recognition, but you can use the same method for face recognition as well.

Just augment your data by GAN and train your model just like this paper method:

https://arxiv.org/ftp/arxiv/papers/1909/1909.11915.pdf

Using only GAN generated data sounds like a bad idea, I don't think the data that be generated will be representative enough. Where GANs are often successfuly used is in augmenting datasets to add more skin tones, ethnicities and races to mitigate overfitting because the dataset creation process is often very sensitive to the cultural context it is performed.