1st Place Solution for ICDAR 2021 Competition on Mathematical Formula Detection

TL;DR

This paper presents a top-performing solution for the ICDAR 2021 mathematical formula detection challenge, utilizing advanced anchor-free detection methods and various tricks to handle scale variation and complex expressions.

Contribution

The authors introduce an effective combination of GFL, ATSS, FPN, and additional techniques like DCN, SyncBN, and WBF for improved mathematical formula detection.

Findings

Achieved 1st place in the ICDAR 2021 MFD competition.

Demonstrated the effectiveness of anchor-free methods and scale handling strategies.

Validated the benefit of tricks like DCN, SyncBN, and WBF in detection performance.

Abstract

In this technical report, we present our 1st place solution for the ICDAR 2021 competition on mathematical formula detection (MFD). The MFD task has three key challenges including a large scale span, large variation of the ratio between height and width, and rich character set and mathematical expressions. Considering these challenges, we used Generalized Focal Loss (GFL), an anchor-free method, instead of the anchor-based method, and prove the Adaptive Training Sampling Strategy (ATSS) and proper Feature Pyramid Network (FPN) can well solve the important issue of scale variation. Meanwhile, we also found some tricks, e.g., Deformable Convolution Network (DCN), SyncBN, and Weighted Box Fusion (WBF), were effective in MFD task. Our proposed method ranked 1st in the final 15 teams.