Towards Learning Object Detectors with Limited Data for Industrial Applications

In this dissertation, three novel Generalized Few-Shot Object Detection (G-FSOD) approaches are presented to minimize the forgetting of previously learned classes while learning new classes with limited data. The first two approaches reduce the forgetting of base classes if they are still available during training. The third approach, for scenarios without base data, uses knowledge distillation to improve the knowledge transfer.

Rapidgator
vz28w.7z.html
Fileaxa
vz28w.7z
Fikper
vz28w.7z.html

​

Click to expand...