SynthTIGER: Synthetic Text Image GEneratoR Towards Better Text Recognition Models

For successful scene text recognition (STR) models, synthetic image generators have alleviated the lack of annotated images from real world. Specifically, they generate multiple with diverse backgrounds, font styles, and shapes enable STR models to learn visual patterns that might not be accessible manually data. In this paper, we introduce a new generator, SynthTIGER, by analyzing techniques used for synthesis integrating effective ones under single algorithm. Moreover, propose two alleviate long-tail problem in length character distributions training our experiments, SynthTIGER achieves better performance than combination datasets, MJSynth (MJ) SynthText (ST). Our ablation study demonstrates benefits using sub-components guideline on generating models. implementation is publicly available at https://github.com/clovaai/synthtiger.