This project is not in working condition

NRTR

A TensorFlow implementation of NRTR, a No-Recurrence Seq2Seq Model for Scene Text Recognition.

It is a Transformer-based model adapted to work with Images as input instead of text.

Paper

For the original paper, see https://arxiv.org/pdf/1806.00926v1.pdf

For the original Transformer architecture paper, see https://arxiv.org/pdf/1706.03762v5.pdf