Great examples of Deep learning?

Hi All,

I’m looking for some great examples of Deep learning applied specifically to vision.

Any real world examples out there that demonstrate the power of AI deep learning?

Facial recognition in facebook picture tagging.
Computer vision in self driving cars.

Would be great to get examples of large scale commercial projects that are in production and ideally profitable.

Anything else?

Thanks

1 Like

See the categories of paperwithcode and there are around 600 vision categories.

That’s great for research. I’d love to see some example of commercial applications.

Any examples of revenue generating businesses using AI for vision?

There are many indirect ways in which vision increases revenue. The phones that are advertised as camera with AI technology, are using vision models underneath it. Check this link. I am not able to think of direct examples right now (3am in my time zone)

Every major AI platform has a vision product (Amazon, Microsoft, Google, and specialists like Clarifai). Have a look at their case studies.

2 Likes

Here are two of my favorites:

Identifying facial phenotypes of genetic disorders using deep learning:
https://www.nature.com/articles/s41591-018-0279-0

Deep learning of aftershock patterns following large earthquakes:
https://www.nature.com/articles/s41586-018-0438-y

1 Like

Identifying facial phenotypes of genetic disorders using deep learning

That’s more of a dystopia scenario, not something desirable.

Every algorithm is depending on its use. In this case, if based on the classification a doctor can be supported on how to help without very expensive blood tests, why not? I don’t see your point, as nowadays every disorder is diagnosed as well, but with different approaches.

Just judging based on the algorithm, I think it is very impressive, what can be inferenced only based on an image from a face. But just my two cents.

The problem is that you will be able to screen people without their consent. It’s easy to use a mobile phone and point it at someone, or use surveillance cams. A blood test is much less likely to happen without consent.

Privacy is a big topic, especially when it comes to health and insurance, because it can have serious real life consequences. From employers using it covertly, to health insurances who will automatically score you and make you pay more or refuse to insure you.

Finally, the blood test is likely to be more reliable, than an even more indirect method based on visual appearance alone.

1 Like

Unfortunately, everything can and will be misused, that is true. But that is a problem of society, and for such kind of problems we do not even need technology :frowning:

Sure, but in this case, I don’t see a big need, compared to just having a blood test.

Just a little insight here from a medical practitioner: the syndromes they are referring to are typically severe genetic syndromes with stereotypical facial features. The problem is that pediatric geneticists are in short supply, and it takes a ton of experience to be able to reliably recognize the large number of genetic diseases, such that a generalist pediatrician can not be aware of all the possible diagnoses (though they certainly know many!). The computer vision techniques here are not really detecting subtle cues that you will, for example, have some kind of late-onset disease that could be used against you in an insurance setting, but are rather helping with early recognition of childhood disorders that need the help of a specialist.

One example they use is Noonan syndrome - check it out, I think you’ll see what I mean and why it would be useful to have CV help.

2 Likes