Robust Open-Vocabulary Translation from Visual Text Representations

Pretrained translation models using visual text representations. 
Models can be interacted with on the command line or through a script similarly to other fairseq models, but do require our code extension for visual text. 
Each zip file contains: dict files, the fairseq model checkpoint, and a sentencepiece model for the target language (English). 

For more information, see the:

• Paper describing the method and training data  [ACL anthology] [arXiv]
• Code repository with scripts  [github]