Information about Cost function in TLT

[KGerry]

Can anyone explain, or redirect me to information about cost function in the TRAIN specfile?
The TLT documentation isn’t very informative about cost function in particular.

Here’s the cost functio part of the detectnet_v2 examplefile:

cost_function_config {
target_classes {
name: “car”
class_weight: 1.0
coverage_foreground_weight: 0.05
objectives {
name: “cov”
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: “bbox”
initial_weight: 10.0
weight_target: 10.0

Can someone tell me what the COV initial weight and target weight mean, same as for the BBOX initial and target weights?

Another question: What should be the settings when training a single class detector?

Thanks alot! Gerard

[Morganh]

Refer to

Tlt spec file - cost function

Hi Morganh, I am not talking about the actual cost function used to train the model itself, but the parameters specified in the spec file I want to know the meaning of the various parameters as the documentation is vague about them. namely - class_weight, coverage_foreground_weight, initial_weight, weight_target etc. Since I need to add more classes to the the spec file(as mentioned above), I wish to have more clarity regarding these parameters as to what value should I give for the nee classe…

Details on cost_function_config for PeopleNet

Refer to https://docs.nvidia.com/metropolis/TLT/tlt-getting-started-guide/text/faqs.html Distribute the dataset class: How do I balance the weight between classes if the dataset has significantly higher samples for one class versus another? To account for imbalance, increase the class_weight for classes with fewer samples. You can also try disabling enable_autoweighting; in this case initial_weight is used to control cov/regression weighting. It is important to keep the number of samples of …

For training a single class detector, it is easy. You just need to delete all the classes you do not want to train.

[KGerry]

Ok, to be really sure, for a single class detector, should I use settings (A)

}
objectives {
  name: "bbox"
  initial_weight: 10.0
  weight_target: 10.0
}

}

Or settings (B)

}
objectives {
  name: "bbox"
  initial_weight: 1.0
  weight_target: 1.0
}

}

[Morganh]

You can set

objectives {
  name: "cov"
  initial_weight: 1.0
  weight_target: 1.0
}
objectives {
  name: "bbox"
  initial_weight: 10.0
  weight_target: 10.0
}