KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Default Boxes
Predefined bounding boxes at different scales and aspect ratios used as anchors to detect objects in an image with SSD.
Multi-scale Feature Layers
Feature maps at different spatial resolutions extracted from the base network to detect objects of various sizes in the image.
Predictive Convolution
3x3 convolutional layers applied to feature maps to simultaneously predict object locations and class scores.
Match Strategy
Association strategy that links ground truth boxes to default boxes based on IoU threshold for supervised training.
Box Regression
Process of predicting precise bounding box coordinates by adjusting default boxes with learned offsets.
Softmax Classification
Activation function applied to confidence scores to produce class probabilities for each detected box.
Multiple Aspect Ratios
Use of default boxes with different width/height ratios (1:1, 2:1, 3:1, etc.) to effectively detect various object shapes.
VGG-16 Backbone
Pre-trained convolutional neural network used as the main feature extractor in the original SSD architecture.
MobileNet SSD
Optimized variant of SSD using MobileNet as backbone for mobile applications with computational constraints.