Frequently Asked Questions

Q: Can a HUNTR scene contain more than one model?

A: Scenes and models are independent of each other. Although models are trained on training areas cut from scenes, the models created can be run on any imagery. See Labeling and Training Best Practices, for more information on best labeling practices.

---

Q: How can I fine tune my model?

A: After creating a model that can recognize the desired class(es) in the imagery, it is helpful to find places that the model is struggling. A good way to fnd these areas is by viewing and downloading the entropy map that HUNTR provides and creating more labels in areas of high entropy.

---

Q: What should I avoid in models with multiple classes?

A:

• Avoid overlapping classes (e.g., airplanes and airports).
• Avoid sparsely labeling one class and generously labeling another (e.g., 1,000 labels for class A and 35 labels for class B).