SoftCTC -- Semi-Supervised Learning for Text Recognition using Soft Pseudo-Labels

TL;DR

SoftCTC introduces a novel semi-supervised learning loss for sequence tasks that considers multiple transcription variants simultaneously, eliminating the need for confidence filtering and improving efficiency.

Contribution

The paper proposes SoftCTC, a new loss function for semi-supervised sequence learning that handles multiple transcriptions without confidence filtering, enhancing efficiency and performance.

Findings

SoftCTC matches the performance of filtered pipelines in handwriting recognition.

It is significantly more computationally efficient than naive CTC approaches.

The GPU implementation is publicly available.

Abstract

This paper explores semi-supervised training for sequence tasks, such as Optical Character Recognition or Automatic Speech Recognition. We propose a novel loss function $\unicode{x2013}$ SoftCTC $\unicode{x2013}$ which is an extension of CTC allowing to consider multiple transcription variants at the same time. This allows to omit the confidence based filtering step which is otherwise a crucial component of pseudo-labeling approaches to semi-supervised learning. We demonstrate the effectiveness of our method on a challenging handwriting recognition task and conclude that SoftCTC matches the performance of a finely-tuned filtering based pipeline. We also evaluated SoftCTC in terms of computational efficiency, concluding that it is significantly more efficient than a na\"ive CTC-based approach for training on multiple transcription variants, and we make our GPU implementation public.