Search for a command to run...
State-of-the-art scene text detection methods usually model the text instance with local pixels or components from the bottom-up perspective and, therefore, are sensitive to noises and dependent on the complicated heuristic post-processing especially for arbitrary-shape texts. To relieve these two issues, instead, we propose to progressively evolve the initial text proposal to arbitrarily shaped text contours in a top-down manner. The initial horizontal text proposals are generated by estimating the center and size of texts. To reduce the range of regression, the first stage of the evolution predicts the corner points of oriented text proposals from the initial horizontal ones. In the second stage, the contours of the oriented text proposals are iteratively regressed to arbitrarily shaped ones. In the last iteration of this stage, we rescore the confidence of the final localized text by utilizing the cues from multiple contour points, rather than the single cue from the initial horizontal proposal center that may be out of arbitrary-shape text regions. Moreover, to facilitate the progressive contour evolution, we design a contour information aggregation mechanism to enrich the feature representation on text contours by considering both the circular topology and semantic context. Experiments conducted on CTW1500, Total-Text, ArT, and TD500 have demonstrated that the proposed method especially excels in line-level arbitrary-shape texts. Code is available at https://github.com/dpengwen/PCR.