Supplementary Materials for AE TextSpotter

To keep the objectivity of TDA-ReCTS validation set, we designed two rules to select ambiguous images from the training set of IC19-ReCTS [3], and then randomly sample 1,000 images among them as the validation set. For an image, we consider its text lines as L, and the internal characters of the ith text line as C in. An image is regarded as a sample with large character spacing, if at least one text line in the image has large character spacing. The character spacing of a text line `i ∈ L is large, if it satisfies Eqn. 1. ∑ cj∈C in mink 6=j,ck∈C inDc(c i j , c i k) ∑ cj∈C in Scale(cj) > 2, (1)