Sequence-to-Sequence Contrastive Learning for Text Recognition

Abstract

We propose a framework for sequence-to-sequence contrastive learning (SeqCLR) of visual representations, which we apply to text recognition. To account for the sequenceto-sequence structure, each feature map is divided into different instances over which the contrastive loss is computed. This operation enables us to contrast in a sub-word level, where from each image we extract several positive pairs and multiple negative examples. To yield effective visual representations for text recognition, we further suggest novel augmentation heuristics, different encoder architectures and custom projection heads. Experiments on handwritten text and on scene text show that when a text decoder is trained on the learned representations, our method outperforms non-sequential contrastive methods. In addition, when the amount of supervision is reduced, SeqCLR significantly improves performance compared with supervised training, and when fine-tuned with 100% of the labels, our method achieves state-of-the-art results on standard handwritten text recognition benchmarks.

Data Augmentation for Scene Text Recognition

“…Typical augmentation methods used include rotation, perspective and affine transformations, Gaussian noise, motion blur, resizing and padding, random or learned distortions, sharpening and cropping [39,23,24,28,1]. Proponents of STR methods select a subset of these augmentation methods to improve their models.…”