KI-Glossar

📖

Begriffe

R-CNN

Two-stage object detection architecture using candidate regions and convolutional neural networks to classify each region. This is a revolutionary approach that significantly improved object detection accuracy in images.

📖

Begriffe

Region of Interest

Rectangular area of the image likely to contain an object, identified during the region proposal phase. These RoIs are then processed individually for classification and localization refinement.

📖

Begriffe

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.