R-CNN (Region-based Convolutional Neural Networks)

Hard Example Mining

Training strategy preferentially selecting difficult examples with high losses to improve detector performance. Particularly effective in R-CNN for managing the imbalance between positive and negative examples.

← Quay lại