Are transformers better at all use cases? Or only a specific subset? This sounds like a classification problem to me, which neural networks have always been pretty good at.
It's mostly a scale thing. Transformers scale better when you have a lot of training data. (Not saying CNNs don't scale, but it's way more difficult to tune the hyper parameters).
For smaller training data sizes, transformers don't have a clear advantage over CNNs. CNNs are still the way to go if you have problems that are not too complex, and especially constraints regarding inference costs. Think of a small edge device / sensor that counts apples in your apple picking machine.
What we sometimes see is that a backbone is trained using transformers, then a smaller CNN is destilled from that (DinoV3 for example).
They have sort of been replaced by transformers for a lot of use cases (and where dataset size allows).
Are transformers better at all use cases? Or only a specific subset? This sounds like a classification problem to me, which neural networks have always been pretty good at.
It's mostly a scale thing. Transformers scale better when you have a lot of training data. (Not saying CNNs don't scale, but it's way more difficult to tune the hyper parameters).
For smaller training data sizes, transformers don't have a clear advantage over CNNs. CNNs are still the way to go if you have problems that are not too complex, and especially constraints regarding inference costs. Think of a small edge device / sensor that counts apples in your apple picking machine.
What we sometimes see is that a backbone is trained using transformers, then a smaller CNN is destilled from that (DinoV3 for example).