Abstract: Highlights•Existing logo detection methods have restricted scalability for new logo classes.•Existing synthesising training data suffers from domain shift limitation.•We present a novel approach for progressive accommodation to new logo classes.
Loading