Context-Free TextSpotter for Real-Time and Mobile End-to-End Text Detection and Recognition

In the deployment of scene-text spotting systems on mobile platforms, lightweight models with low computation are preferable. concept, end-to-end (E2E) text is suitable for such purposes because it performs detection and recognition in a single model. However, current state-of-the-art E2E methods rely heavy feature extractors, recurrent sequence modellings, complex shape aligners to pursue accuracy, which means their computations still heavy. We explore opposite direction: How far can we go without bells whistles spotting? To this end, propose text-spotting method that consists simple convolutions few post-processes, named Context-Free TextSpotter. Experiments using standard benchmarks show TextSpotter achieves real-time GPU only three million parameters, smallest fastest among existing deep spotters, an acceptable transcription quality degradation compared heavier ones. Further, demonstrate our spotter run smartphone affordable latency, valuable building stand-alone OCR applications.