Box2Poly: Memory-Efficient Polygon Prediction of Arbitrarily Shaped and Rotated Text
This work addresses memory and speed bottlenecks in text detection for applications like document analysis and scene text recognition, representing an incremental improvement over existing methods.
The paper tackles the problem of memory inefficiency and poor vertex relationship capture in Transformer-based text detection by proposing Box2Poly, a cascade decoding pipeline based on Sparse R-CNN for polygon prediction of arbitrarily shaped and rotated text. It achieves over 50% memory reduction and over 40% faster inference compared to the state-of-the-art DPText-DETR, with only a minor performance drop on benchmarks.
Recently, Transformer-based text detection techniques have sought to predict polygons by encoding the coordinates of individual boundary vertices using distinct query features. However, this approach incurs a significant memory overhead and struggles to effectively capture the intricate relationships between vertices belonging to the same instance. Consequently, irregular text layouts often lead to the prediction of outlined vertices, diminishing the quality of results. To address these challenges, we present an innovative approach rooted in Sparse R-CNN: a cascade decoding pipeline for polygon prediction. Our method ensures precision by iteratively refining polygon predictions, considering both the scale and location of preceding results. Leveraging this stabilized regression pipeline, even employing just a single feature vector to guide polygon instance regression yields promising detection results. Simultaneously, the leverage of instance-level feature proposal substantially enhances memory efficiency (>50% less vs. the state-of-the-art method DPText-DETR) and reduces inference speed (>40% less vs. DPText-DETR) with minor performance drop on benchmarks.