WeCromCL: Weakly Supervised Cross-Modality Contrastive Learning for Transcription-only Supervised Text Spotting

TL;DR

This paper introduces WeCromCL, a weakly supervised contrastive learning approach that detects text transcriptions in images without boundary annotations, significantly reducing annotation costs while maintaining high accuracy.

Contribution

WeCromCL is the first to apply atomistic contrastive learning for weakly supervised text spotting, modeling character-wise appearance consistency for localization.

Findings

Outperforms existing methods on four benchmarks.

Effectively detects transcriptions without boundary annotations.

Demonstrates robustness across challenging scene text datasets.

Abstract

Transcription-only Supervised Text Spotting aims to learn text spotters relying only on transcriptions but no text boundaries for supervision, thus eliminating expensive boundary annotation. The crux of this task lies in locating each transcription in scene text images without location annotations. In this work, we formulate this challenging problem as a Weakly Supervised Cross-modality Contrastive Learning problem, and design a simple yet effective model dubbed WeCromCL that is able to detect each transcription in a scene image in a weakly supervised manner. Unlike typical methods for cross-modality contrastive learning that focus on modeling the holistic semantic correlation between an entire image and a text description, our WeCromCL conducts atomistic contrastive learning to model the character-wise appearance consistency between a text transcription and its correlated region in a scene image to detect an anchor point for the transcription in a weakly supervised manner. The detected anchor points by WeCromCL are further used as pseudo location labels to guide the learning of text spotting. Extensive experiments on four challenging benchmarks demonstrate the superior performance of our model over other methods. Code will be released.