Weak Supervision for Generating Pixel-Level Annotations in Scene Text Segmentation

TL;DR

This paper introduces a weakly supervised learning method for scene text segmentation that reduces reliance on synthetic data by leveraging bounding-box annotations to generate pixel-level labels, improving performance with less data.

Contribution

The authors propose a novel weak supervision approach using bounding-box annotations to generate pixel-level labels, along with the creation of new datasets and a specialized CNN architecture for scene text segmentation.

Findings

Weak supervision reduces the need for extensive pixel-level annotations.

Generated datasets improve segmentation performance significantly.

The proposed method outperforms synthetic data pre-training approaches.

Abstract

Providing pixel-level supervisions for scene text segmentation is inherently difficult and costly, so that only few small datasets are available for this task. To face the scarcity of training data, previous approaches based on Convolutional Neural Networks (CNNs) rely on the use of a synthetic dataset for pre-training. However, synthetic data cannot reproduce the complexity and variability of natural images. In this work, we propose to use a weakly supervised learning approach to reduce the domain-shift between synthetic and real data. Leveraging the bounding-box supervision of the COCO-Text and the MLT datasets, we generate weak pixel-level supervisions of real images. In particular, the COCO-Text-Segmentation (COCO_TS) and the MLT-Segmentation (MLT_S) datasets are created and released. These two datasets are used to train a CNN, the Segmentation Multiscale Attention Network (SMANet), which is specifically designed to face some peculiarities of the scene text segmentation task. The SMANet is trained end-to-end on the proposed datasets, and the experiments show that COCO_TS and MLT_S are a valid alternative to synthetic images, allowing to use only a fraction of the training samples and improving significantly the performances.