Automating the back office of banks is essential to improving financial access. However, just automating what humans do today, does not solve the problem completely.
Synapse’s current solutions (ID Verification and Video Auth) work well for people with a good camera and internet. But we want to build a platform that works for all, so we need to do better.
Over the last few months we have been working on some Generative Adversarial Networks (GANs) that will help us reconstruct facial features from fairly noisy images. We call it FR-GAN (Facial Reconstruction GAN; pronounced free·gan).
It’s very critical to build these networks because not everyone has a high quality camera. And with this, we can broaden the scope of people who can have access to quality banking, regardless of camera quality of their phones.
Over the coming weeks, FR-GAN will start powering our ID Verification and Video Auth tools.
Here are some results:
All the images on the right are reconstructed from the images on the left using our FR-GAN.¹
The last correction is a little nutty because it takes a very blurry image and reconstructs a fairly good facial model. We are probably not going to allow that on production because it’s a little extreme (since it’s illegible to humans). But it does demonstrate how powerful the network really is.
If you want to build tools like these and are really passionate about building a financial system that works for all, come join our team.
 We are working on improving performance in under or oversaturated light conditions as well.