I am trying to find research paper with theory(preferably implementation) that is about classifying 1000 (or more) classes. I have heard of an implementation, that initially clustering needs to be done then classification with something like softmax. Does anyone know of any research paper that implements 1000+ class classification.
If you are asking for arbitrary ML task dealing with 1000+ classes the most straigtforward thing that comes to mind is the
ImageNet - https://en.wikipedia.org/wiki/ImageNet#cite_note-nytimes_2012-2. It has more the 20k categories in the present time.
In order to perform such classification task you need
- A large enough dataset, such that each class occurs frequently enough or efficient data augmentation
- A powerful model, such that can adapt to variety of different problems
The first neural network to tackle this problem was
AlexNet and since then plethora of architectures :
EfficientNet got better and better quality.
They have a lot of filters that react on different patterns.