Let’s say I want to classify vehicle colors. Are one way better then the other in terms of accuracy or it’s the same?
- Data set containing all classes - red, green, blue … and crate one classification model using this data set.
- separate dateset for each color - (red, other colors)(green, other colors) … and creating classification model for each of the colors/classes?
The first one will be harder for classification task. The more classes the classifier needs to know, the harder job it is.
The second option is a better one, if your task is always like ‘is this car red?’ Such question is often asked when searching for ‘red’ cars on surveillance systems.
However, if your task is to tell the color of a car, you have either go for option 1, or option 2 with another decision to pick the final result.