Detectnet Cost function Config

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

The tlt getting started, does not go into these params in detail - specific information on how they effect model training / performance will be helpful. This is from the cot function config use by detectnet:

The ones in the bold, I am unsure of and the default values given, but I would like to know their usecases exactly so I can have the best accuracy.

target_classes {
name: “cardboard”
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: “cov”
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}

2

Refer to

https://docs.nvidia.com/metropolis/TLT/FAQ/index.html

1 Like
3

It does not explain how it affects training or on what concrete heuristics I should even define the value. It is extremely vague and provides no detail to me how I should set the params for my custom dataset:

How should I set class weight based off my custom dataset?
The same thing for initial weight and weight target? Right now it feels I am just using random values without any intelligent context behind those decisions.

4

The class weight is assigned to the corresponding target class’s cost. It can balance the training dataset for each class. Weights 10, 8, 6 are the same as 5, 4, 3 since the proportion of each class is the same for the two cases.

Some other tips from FAQ.

  • 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 different classes balanced, which helps improve mAP.

  • In DetectNet_V2, are there any parameters that can help improve AP (average precision) on training small objects? Following parameters can help you improve AP on smaller objects:

  • Increase num_layers of resnet

  • class_weight for small objects

  • Increase the coverage_radius_x and coverage_radius_y parameters of the bbox_rasterizer_config section for the small objects class

  • Decrease minimum_detection_ground_truth_overlap

  • Lower minimum_height to cover more small objects for evaluation.

1 Like