- Task 1 - Text Localization
- Task 2 - Script identification
- Task 3 - Joint text detection and script identification
method: 4Paradigm-Data-Intelligence2019-06-03
Authors: Feng Cheng, Lixin Gu, Qingjie Liu, Feng Han, Jingtao Han
Description: The detection model and recognition model are trained separately.
Detection model: Based on Mask-RCNN. multi-scale. Train-set: 2017 MLT task1 train-set.
Recognition model: Based on Transformer with backbone ResNet50. A voting process is done to identify the language of recognized transcript. Train-set: 2017 MLT task2 train-set & 2019 MLT task2 train-set & 2019 MLT Synthetic dataset.
method: PMTD + CNN based method2019-09-28
Authors: Geonho Hwang
Affiliation: NCIA, Seoul National University
Description: Task1: PMTD
Task2: CNN based method
method: CLOVA-AI2019-06-04
Authors: Bado Lee, Youngmin Baek, Hwalsuk Lee
Description: Additional head on Character-level text detection with model distillation. A pretrained detector is used.
CLOVA-AI team, Naver Corp.
| Date | Method | Hmean | Precision | Recall | Average Precision | |||
|---|---|---|---|---|---|---|---|---|
| 2019-06-03 | 4Paradigm-Data-Intelligence | 75.23% | 79.26% | 71.60% | 56.65% | |||
| 2019-09-28 | PMTD + CNN based method | 72.64% | 78.76% | 67.39% | 61.48% | |||
| 2019-06-04 | CLOVA-AI | 68.31% | 74.52% | 63.06% | 54.56% | |||
| 2019-05-10 | Ashwaq | 58.11% | 62.44% | 54.34% | 40.61% | |||
| 2017-06-30 | SCUT-DLVClab2 | 58.08% | 71.78% | 48.77% | 41.42% | |||
| 2017-06-30 | TH-DL | 39.37% | 58.58% | 29.65% | 24.54% |