Fig. 4From: An end-to-end text spotter with text relation networksThe numbers of text instances contained in the images in the three datasetsBack to article page