Tag
This paper introduces a large-scale synthetic dataset (WATER-S) and a specialized model (WATERec) to advance WordArt-oriented scene text recognition, achieving state-of-the-art accuracy on irregular artistic text benchmarks.
Raster2Seq reconstructs floorplan vector graphics from raster images using a sequence-to-sequence approach with autoregressive decoding guided by learnable anchors, achieving state-of-the-art performance on multiple benchmarks.