Tensorflow object-detection API Num class

Question

I would like to know if the values of num_classes in Tensorflow object-detection API can affects the model performance. For example , lets say I am training the model to detect only one object(say human) but in the the num_classes I set it 20. Does incorrect number of num_classes will have affects on the model performance?

Answer 1

It's not critical if you set num_classes to be higher than the actual number of classes (it only makes some tensors/variables a bit larger), but it is critical if you set it to be lower. the first case shouldn't affect the model performance.