[Master Thesis(Research)] Enhance data selection efficiency with variational auto-encoder for object detection’s active learning
Abstract:
• It can save 70% labeled cost for achieving a usable model in diverse domains and rare events.
• Uncertainty and diversity information are important for active learning.
• VAE can provide good representation for known the unlabeled data distribution in surveillance cameras.
Reference:
Document:
https://etd.lib.nctu.edu.tw/cgi-bin/gs32/hugsweb.cgi?o=dnthucdr&s=id=%22G021060617050%22.&searchmode=basic