It really depends. As you read the code, the inspiration is that, we assume the normal feature are well distributed (clustered). If more backgrounds are included, you can think of more clusters are added. Ideally, if those added normal feature's clusters are still stay away from each other, that might not be a big problem.
It really depends. As you read the code, the inspiration is that, we assume the normal feature are well distributed (clustered). If more backgrounds are included, you can think of more clusters are added. Ideally, if those added normal feature's clusters are still stay away from each other, that might not be a big problem.