TS-Net: OCR Trained to Switch Between Text Transcription Styles

Users of OCR systems, from different institutions and scientific disciplines, prefer produce transcription styles. This presents a problem for training consistent text recognition neural networks on real-world data. We propose to extend existing with Transcription Style Block (TSB) which can learn data switch between multiple styles without any explicit knowledge rules. TSB is an adaptive instance normalization conditioned by identifiers representing consistently transcribed documents (e.g. single document, transcriber, or institution). show that able completely in controlled experiments artificial data, it improves accuracy large-scale learns semantically meaningful style embeddings. also how efficiently adapt new transcriptions only few lines.