logo SBA

ETD

Digital archive of theses discussed at the University of Pisa

 

Thesis etd-03262025-221251

Link copiato negli appunti
Thesis type
Tesi di laurea magistrale
URN
etd-03262025-221251
Thesis title
Layer-wise Approach to Activation Analysis in YOLOv8 Model: Measuring Localization Effectiveness through GradCAM
Department
INFORMATICA
Course of study
INFORMATICA
Keywords
  • density
  • interpretation
  • YOLO
Graduation session start date
11/04/2025
Availability
Withheld
Release date
11/04/2028
Abstract (Inglese)
Abstract (Italiano)
This research analyzes activation patterns across YOLOv8's three scale-specific convolutional layers using GradCAM. We measure activation density and top-k pixel distribution to quantify detection effectiveness for small, medium, and large objects. Our novel scale classification model determines object size based on layer-specific
activation patterns, providing insights into YOLO's multi-scale detection mechanisms
File
Nome file Dimensione
The thesis is not available.
Contact the author