Multilingual semantic fusion network for text recognition in the wild

MSFN is a pure Transformer model for scene text recognization. It uses a Vision Transformer for image feature extraction and a multi-lingual transformer decoder for text generation.