Glosarium AI
Kamus lengkap Kecerdasan Buatan
Mean Average Precision (mAP)
Standard evaluation metric in object detection that calculates the mean of average precisions across all classes, taking into account both precision and recall.
SSD (Single Shot Detector)
Object detection architecture that predicts bounding boxes and classes in a single pass using multi-scale feature maps.
Object Localization
Computer vision task that involves identifying the presence and precise position of specific objects in an image, typically with bounding boxes.
Confidence Score
A numerical value between 0 and 1 indicating the model's degree of certainty that a given detection actually contains an object of the predicted class.
Ground Truth
Precise manual annotations of bounding boxes and object classes in a training dataset, serving as a reference for supervised learning.
Multi-Scale Detection
Detection strategy that processes the image at different resolutions to improve the detection of objects with highly varied sizes, from very small to very large.
Hard Negative Mining
Training technique that selects and emphasizes difficult negative examples (false positives with high scores) to improve the model's discrimination.
Focal Loss
Modified loss function that addresses class imbalance by penalizing misclassified and easy examples more heavily, thereby improving detection of rare objects.
Region of Interest (RoI) Pooling
Operation that extracts fixed-size features from variable-size regions in feature maps, essential for two-stage detectors.