method: BreSee OCR2021-07-01
Authors: Mengyue Shao, Jie Wu, Jiling Wu, Tianpeng Li, Linzhi Zhuang
Affiliation: BreSee AI Lab, Zhejiang Sci-Tech University
Description: In this task1, we follow YOLOv5 and DBNet as the base model to do detection task. And some targeted optimization of models have been carried out according to the data set provided, and better results have been achieved. The two basic models are fused to reach the final result.
method: Samsung Life Insurance2020-10-16
Authors: Dongyoung Kim, Myungsung Kwak
Affiliation: Data Analytics Laboratory (DA Lab), Samsung Life Insurance
Description: A document Text Localization Generative Adversarial Nets (TLGAN) model is utilized to perform the text localization task using SROIE data set. TLGAN learns text-image features via ImageNet pre-trained VGG network in adversarial manner and points out text locations. Note the images were scaled in an arbitrary ratio and the detected coordinates were re-scaled into original image space for the submission.
method: Sunshine_OCR2021-03-10
Authors: Sunshine AI Team
Affiliation: Sunshine Insurance Group
Description: We use rot-faster-rcnn/rot-mask-rcnn algorithm which adds rotated rectangle support based on faster-rcnn/mask-rcnn. Firstly ResNet50 backbone + rot-faster-rcnn is used to get a 98.5%+ baseline, then HRNet backone + rot-mask-rcnn +IOU loss + multiscale training&testing is used to get the final score. The data is preprocessed for scale alignment and only augmented by small rotations.
Note: No error fixing/excluding is done on the submit.
Date | Method | Recall | Precision | Hmean | |||
---|---|---|---|---|---|---|---|
2021-07-01 | BreSee OCR | 99.15% | 99.40% | 99.28% | |||
2020-10-16 | Samsung Life Insurance | 98.64% | 99.83% | 99.23% | |||
2021-03-10 | Sunshine_OCR | 98.81% | 98.97% | 98.89% | |||
2020-08-10 | BOE_AIoT_CTO | 98.76% | 98.92% | 98.84% | |||
2019-04-22 | H&H Lab | 97.93% | 97.95% | 97.94% | |||
2022-05-09 | A modified CTPN model 2.0 | 97.52% | 97.40% | 97.46% | |||
2021-10-22 | A modified CTPN model 1.0 | 97.16% | 97.10% | 97.13% | |||
2020-09-27 | only PAN | 96.51% | 96.80% | 96.66% | |||
2021-01-28 | 58CV | 97.48% | 95.43% | 96.45% | |||
2019-04-22 | GREAT-OCR Shanghai University | 96.62% | 96.21% | 96.42% | |||
2019-04-23 | BOE_IOT_AIBD | 95.95% | 95.99% | 95.97% | |||
2019-04-21 | IFLYTEK-textDet_v3 | 93.77% | 95.89% | 94.81% | |||
2019-04-22 | A Single-Shot Model for Robust Text Localization | 93.93% | 94.80% | 94.37% | |||
2019-04-19 | BiLSTM Based on CTPN | 91.40% | 94.03% | 92.69% | |||
2019-04-17 | EAST_clip_enhance_896_giou | 89.69% | 93.77% | 91.68% | |||
2019-04-17 | Textline detection | 89.85% | 92.72% | 91.26% | |||
2019-04-20 | A Text Localization Method Based on CTPN | 85.23% | 88.73% | 86.94% | |||
2019-04-16 | Vsdnu | 85.07% | 87.17% | 86.11% | |||
2021-05-10 | Original CRAFT for SROIE | 62.73% | 59.94% | 61.31% | |||
2019-04-17 | scene text detection weapon | 49.61% | 64.75% | 56.18% | |||
2021-04-13 | Practicing project for Scientific Research Subject (HCMUS master program) | 37.02% | 30.07% | 33.19% |