Deep Learning Training Dataset with Caffe

Question

I am a beginner novice and am working on creating a car classifier for images using Caffe and ask a three-part question:

• Are there any recommendations for organizing classes for CNN training? that is, the number of classes and the number of samples for each class? For example, is it better for me to do this:

  • (a) Vehicles - car sedans / car-hatchback / car-SUV / truck-18-wheeler / .... (note that this may mean several thousand classes) or
  • (b) have a higher level model that classifies between car / truck / 2-wheeler and so on ... and if the car type then requests the car model to get the car type
    (sedan / hatchback, etc.)

• How many classroom images are typical best practice? I know that there are several other variables that affect the accuracy of CNN, but what rough number can be removed in each class? Should there be a function of the number of classes in the model? For example, if I have many classes in my model, should I provide more samples in the class?

• How do we guarantee that we do not redefine the class? Is there a way to measure heterogeneity in classroom samples?

Thanks in advance.

Answer 1

• , , , : , , , Car! this, . , , /, CNN, , . , ..

• , , , . . , .

• , CNN . , .

, , , , (, ILSVRC 201X) . ( ) .