Authors: Xingran Zhao, Jing Xian, Yadong Li, Hongbin Wang

Affiliation: AntGroup


Description: For word and line detection, we firstly crop patches from images for catching local mask results. Second, we also get global mask results by using full images as the input. Thirdly, we merge global and local results by using NMS postprocess procedure. For paragraph detection, we only use full images as input and get global mask results. All detectors are CBNetV2[1] with HTC[2]. For hierarchical text detection, we use IOS(intersection-of-sets) as metric to assign words into lines and use same strategy to assign lines into paragraphs.
[1]CBNetV2: A Composite Backbone Network Architecture for Object Detection.
[2]Hybrid Task Cascade for Instance Segmentation.

method: Hi-SAM2023-12-28

Authors: Maoyuan Ye, Jing Zhang, Juhua Liu, Chenyu Liu, Baocai Yin, Cong Liu, Bo Du, Dacheng Tao

Description: A unified text segmentation model across four hierarchies, including stroke, word, text-line, and paragraph, while realizing layout analysis as well. Only the training data of HierText is adopted.

Authors: Shangbang Long, Siyang Qin, Dmitry Panteleev, Alessandro Bissacco, Yasuhisa Fujii, Michalis Raptis

Affiliation: Google Research

Description: This official submission accompanies our paper: Towards End-to-End Unified Scene Text Detection and Layout Analysis. Note that the unified detector model produces line-level masks and a line-wise affinity matrix that groups lines into paragraphs. It is unable to produce word-level detection directly. For evaluation purpose solely, we use heuristics to extract word masks from line masks. Please refer to the source code ( to learn how this is performed.

Ranking Table

Description Paper Source Code
2023-04-01Global and local instance segmentations for hierarchical text detection0.76160.90720.93450.88160.83950.68500.82220.80240.84310.83310.62550.75110.74000.76250.8328
2022-08-09Unified Detector (CVPR 2022 version)0.48210.61510.67540.56470.78380.62230.79910.79640.80190.77870.53600.68580.76040.62450.7817
2023-02-06HierText official ckpt0.47990.61350.67190.56450.78220.62200.79980.80000.79960.77770.53510.68560.76540.62080.7805

Ranking Graphic

Ranking Graphic - Line PQ

Ranking Graphic - Paragraph PQ