Deep Learning Essentials

Others

For different problems, object functions can be different for the purpose of better determining the differences between the estimate versus the ground truth (note that ground truth may not just be a scale-like class label, for example, as in detection problems). For example, for object detection problems, there’s a loss function named focal loss, which adds a factor to the standard cross-entropy criterion to help the learning focus on a sparse set of hard examples and prevent the vast number of easy negatives from overwhelming the detector during training.