method: TextFuseNet2020-07-31
Authors: Jian Ye, Zhe Chen, Juhua Liu and Bo Du
Affiliation: Wuhan University, The University of Sydney
Email: liujuhua@whu.edu.cn
Description: Arbitrary shape text detection in natural scenes is an extremely challenging task. Unlike existing text detection approaches that only perceive texts based on limited feature representations, we propose a novel framework, namely TextFuseNet, to exploit the use of richer features fused for text detection. More specifically, we propose to perceive texts from three levels of feature representations, i.e., character-, word- and global-level, and then introduce a novel text representation fusion technique to help achieve robust arbitrary text detection. The multi-level feature representation can adequately describe texts by dissecting them into individual characters while still maintaining their general semantics. TextFuseNet then collects and merges the texts’ features from different levels using a multi-path fusion architecture which can effectively align and fuse different representations. In practice, our proposed TextFuseNet can learn a more adequate description of arbitrary shapes texts, suppressing false positives and producing more accurate detection results. Our proposed framework can also be trained with weak supervision for those datasets that lack character-level annotations. Experiments on several datasets show that the proposed TextFuseNet achieves state-of-the-art performance. Specifically, we achieve an F-measure of 94.3% on ICDAR2013, 92.1% on ICDAR2015,87.1% on Total-Text and 86.6% on CTW-1500, respectively.
method: TH2020-01-22
Authors: Tsinghua University and Hyundai Motor Group AIRS Company
Email: Shanyu Xiao: xiaosy19@mails.tsinghua.edu.cn
Description: We have built an end-to-end scene text spotter based on Mask R-CNN & Transformer. The ResNet-50 backbone and multiscale training/testing are used.
method: JDAI2019-08-13
Authors: Jingyang Lin, Jiajia Geng, Rongfeng Lai
Description: We are from JDAI and Sun Yat-Sen University. It is a strong scene text detection baseline built upon Mask R-CNN architecture.
Date | Method | Recall | Precision | Hmean | |||
---|---|---|---|---|---|---|---|
2020-07-31 | TextFuseNet | 90.56% | 93.96% | 92.23% | |||
2020-01-22 | TH | 89.46% | 94.03% | 91.69% | |||
2019-08-13 | JDAI | 90.85% | 92.50% | 91.67% | |||
2019-03-11 | Alibaba-PAI V2 | 89.41% | 93.32% | 91.32% | |||
2019-08-08 | Eleme-AI | 89.31% | 93.03% | 91.13% | |||
2018-09-07 | Sogou_MM | 90.03% | 92.21% | 91.11% | |||
2018-07-03 | Baidu VIS v2 | 88.11% | 94.04% | 90.98% | |||
2022-04-26 | NVTextSpotter | 91.33% | 90.03% | 90.68% | |||
2019-05-14 | ArtDet | 88.40% | 92.91% | 90.60% | |||
2018-01-31 | Alibaba-PAI | 87.34% | 93.84% | 90.47% | |||
2018-01-22 | FOTS | 87.92% | 91.85% | 89.84% | |||
2018-11-15 | Pixel-Anchor(Multiscale) | 86.95% | 92.28% | 89.54% | |||
2018-03-05 | HoText_v1 | 83.58% | 96.34% | 89.51% | |||
2019-08-07 | CM-CV&AR | 87.53% | 91.49% | 89.47% | |||
2020-08-12 | RRPN++ (single scale) | 87.19% | 91.84% | 89.45% | |||
2019-04-11 | Shape-Aware Arbitrary Text Detector | 87.77% | 91.20% | 89.45% | |||
2020-09-26 | DCLNet | 87.05% | 90.31% | 88.65% | |||
2017-09-15 | Baidu VIS | 83.39% | 93.62% | 88.21% | |||
2017-12-19 | HIK_OCR_V1 | 84.59% | 91.94% | 88.11% | |||
2019-05-14 | ArtDet (single scale) | 84.93% | 91.40% | 88.05% | |||
2018-06-26 | SPCNet_TongJi & UESTC (single scale) | 86.71% | 88.94% | 87.81% | |||
2017-09-12 | SenseTime V3 | 84.83% | 90.82% | 87.73% | |||
2019-05-17 | SEG-PIXEL-PAN (single-scale) | 85.32% | 90.22% | 87.70% | |||
2018-11-15 | Pixel-Anchor(single scale) | 87.05% | 88.32% | 87.68% | |||
2018-05-18 | PSENet_NJU_ImagineLab (single-scale) | 85.22% | 89.30% | 87.21% | |||
2017-09-11 | Dahua-OCR V3 | 83.44% | 91.31% | 87.19% | |||
2018-12-03 | CV_OCR_NOOB(single-scale) | 82.96% | 91.55% | 87.04% | |||
2020-10-16 | Drew | 84.11% | 90.10% | 87.00% | |||
2019-04-08 | CRAFT | 84.26% | 89.79% | 86.93% | |||
2018-01-16 | HappyCCL | 83.87% | 89.10% | 86.41% | |||
2017-12-25 | Alibaba-PAI | 82.91% | 89.55% | 86.10% | |||
2017-07-12 | Tencent-DPPR | 81.80% | 90.71% | 86.03% | |||
2019-05-30 | Lanm | 84.26% | 87.28% | 85.74% | |||
2017-07-04 | Baidu IDL v3 | 81.99% | 89.82% | 85.73% | |||
2018-07-09 | BUPT ICYBEE | 81.99% | 89.73% | 85.69% | |||
2017-05-13 | SRC-B-MachineLearningLab-v3 | 82.81% | 88.66% | 85.64% | |||
2020-04-18 | MMLab-PolarMask++(Single Scale) | 83.53% | 87.36% | 85.40% | |||
2017-09-13 | PixelLink | 83.77% | 86.65% | 85.19% | |||
2018-12-02 | EPTN-SJTU | 80.93% | 89.13% | 84.83% | |||
2017-07-26 | TextBoxes++ | 80.79% | 89.11% | 84.75% | |||
2018-01-04 | crpn | 80.69% | 88.77% | 84.54% | |||
2017-11-26 | blank_net_v1 | 80.79% | 88.46% | 84.45% | |||
2017-10-22 | FTDN-SJTU-v2 | 80.93% | 87.69% | 84.18% | |||
2022-04-19 | test | 78.48% | 90.71% | 84.15% | |||
2017-09-03 | CCFLAB_FTSN | 80.07% | 88.65% | 84.14% | |||
2017-09-07 | cvte_zju_v2 | 84.40% | 83.64% | 84.02% | |||
2018-05-07 | textDetection | 84.83% | 83.19% | 84.00% | |||
2017-02-17 | NLPR-CASIA | 82.76% | 84.76% | 83.75% | |||
2017-06-11 | SCUT_DMPNet_pro | 83.73% | 83.53% | 83.63% | |||
2018-08-09 | YY-tl_final | 82.43% | 84.63% | 83.51% | |||
2017-09-04 | FTDN-SJTU | 80.55% | 86.59% | 83.46% | |||
2019-07-15 | stela | 78.57% | 88.70% | 83.33% | |||
2019-04-10 | EAST-VGG16 | 81.27% | 84.36% | 82.79% | |||
2018-01-10 | HoText_v0 | 79.20% | 86.40% | 82.64% | |||
2020-08-14 | DAL(multi-scale) | 80.45% | 84.35% | 82.36% | |||
2018-04-28 | train twice | 79.20% | 85.77% | 82.35% | |||
2018-01-08 | q | 80.74% | 83.81% | 82.25% | |||
2017-08-18 | Dahua-OCR v1 | 79.30% | 85.16% | 82.12% | |||
2018-04-26 | east-modified | 78.57% | 85.40% | 81.85% | |||
2017-02-22 | SRC-B-MachineLearningLab-v2 | 79.78% | 83.64% | 81.67% | |||
2019-08-02 | PyTorch re-implementation of EAST | 74.48% | 90.26% | 81.61% | |||
2020-07-12 | east improved | 76.50% | 87.40% | 81.59% | |||
2020-08-13 | DAL | 79.49% | 83.68% | 81.53% | |||
2017-08-14 | MultDet | 78.48% | 84.32% | 81.30% | |||
2017-01-21 | CNN based model | 79.68% | 82.34% | 80.99% | |||
2017-07-31 | EAST reimplemention with resnet 50 | 77.32% | 84.66% | 80.83% | |||
2017-12-11 | Huaibeicun TextDetector | 77.32% | 84.66% | 80.83% | |||
2017-06-19 | Megvii-EAST | 78.33% | 83.27% | 80.72% | |||
2017-10-18 | ICT | 76.55% | 85.21% | 80.65% | |||
2017-07-31 | dengdan_zju_cad | 80.36% | 80.43% | 80.39% | |||
2020-10-21 | Huawei_GDE_AI | 74.63% | 86.88% | 80.29% | |||
2017-08-11 | SegRPN | 75.83% | 85.18% | 80.23% | |||
2017-07-12 | CCFLAB_v1 | 77.03% | 83.68% | 80.22% | |||
2017-01-23 | RRPN-4 | 77.13% | 83.52% | 80.20% | |||
2019-03-08 | R2CNN++ (single scale) | 78.86% | 81.33% | 80.08% | |||
2017-11-21 | ITN_VGG | 74.15% | 85.70% | 79.50% | |||
2017-01-13 | MSRA_v1 | 74.10% | 85.22% | 79.27% | |||
2019-07-11 | AFCTPN | 76.02% | 82.71% | 79.23% | |||
2018-05-07 | vm24 | 74.92% | 82.63% | 78.59% | |||
2018-04-27 | my_east_with_atrous | 75.25% | 81.96% | 78.46% | |||
2020-08-15 | R-RetinaNet | 77.80% | 77.25% | 77.52% | |||
2016-10-28 | RRPN-3 | 73.23% | 82.17% | 77.44% | |||
2017-07-13 | knet0 | 72.46% | 83.06% | 77.40% | |||
2017-01-19 | SRC-B-MachineLearningLab | 69.86% | 86.11% | 77.14% | |||
2017-01-13 | SCUT_DMP_v2 | 76.89% | 77.00% | 76.95% | |||
2017-02-12 | SSTD | 73.86% | 80.23% | 76.91% | |||
2017-07-13 | knet1 | 76.89% | 76.01% | 76.45% | |||
2017-07-14 | zju_cvte_seglink_512 | 72.85% | 80.22% | 76.36% | |||
2017-07-09 | CCFLAB | 75.06% | 77.25% | 76.14% | |||
2018-08-06 | YY_tl_v1 | 69.23% | 83.65% | 75.76% | |||
2017-01-14 | hust_orientedText | 76.50% | 74.74% | 75.61% | |||
2017-10-12 | DDR_v4 | 73.57% | 77.76% | 75.61% | |||
2022-01-24 | another_segText | 74.00% | 76.62% | 75.29% | |||
2016-10-25 | Baidu IDL v2 | 72.75% | 77.41% | 75.01% | |||
2017-10-24 | BJUT_FNIC | 70.73% | 79.45% | 74.83% | |||
2017-05-01 | TsinghuaOCR | 67.74% | 83.21% | 74.68% | |||
2017-10-12 | DDR_v3 | 74.96% | 72.02% | 73.46% | |||
2023-05-06 | exp2 | 67.69% | 80.16% | 73.40% | |||
2023-05-06 | exp4 | 67.02% | 78.20% | 72.18% | |||
2017-10-12 | DDR_v2 | 74.19% | 69.32% | 71.67% | |||
2023-05-06 | exp1 | 65.38% | 79.05% | 71.57% | |||
2016-09-23 | HUST_OrientedText | 74.15% | 69.03% | 71.49% | |||
2016-10-25 | SCUT_DMPNet | 68.22% | 73.23% | 70.64% | |||
2016-06-23 | Baidu IDL | 68.22% | 71.53% | 69.84% | |||
2023-05-06 | exp3 | 63.99% | 76.07% | 69.51% | |||
2018-07-27 | YY-tl | 67.21% | 67.80% | 67.50% | |||
2015-11-11 | Megvii-Image++ | 56.96% | 72.40% | 63.76% | |||
2017-10-12 | DDR_v1 | 72.80% | 55.98% | 63.29% | |||
2019-07-23 | std++(single-scale) | 56.67% | 71.64% | 63.28% | |||
2020-05-04 | RESULTS | 66.39% | 60.30% | 63.20% | |||
2016-11-08 | CTPN | 51.56% | 74.22% | 60.85% | |||
2018-04-27 | erTree_mxFlow_sigleDir | 55.32% | 66.07% | 60.22% | |||
2018-12-27 | fast_ret_sh_02 | 54.07% | 65.87% | 59.39% | |||
2017-09-21 | UCAS_CMVT3 | 49.16% | 66.38% | 56.49% | |||
2020-09-05 | zoomtext | 47.76% | 68.65% | 56.33% | |||
2016-01-15 | MCLAB_FCN | 43.09% | 70.81% | 53.58% | |||
2019-01-27 | rrpn ssprpns | 50.79% | 50.55% | 50.67% | |||
2017-09-08 | CMVT_resNet | 42.99% | 60.26% | 50.18% | |||
2015-04-03 | Stradvision-2 | 36.74% | 77.46% | 49.84% | |||
2019-01-27 | rrpn example | 50.02% | 49.43% | 49.72% | |||
2018-03-29 | resNet-32r-3dp | 46.65% | 52.95% | 49.60% | |||
2015-04-02 | StradVision-1 | 46.27% | 53.39% | 49.57% | |||
2015-12-15 | CASIA_USTB-Cascaded | 39.53% | 61.68% | 48.18% | |||
2015-04-02 | NJU_Text_Version4 | 35.82% | 72.73% | 48.00% | |||
2015-04-01 | NJU Text (Version2) | 36.25% | 70.44% | 47.87% | |||
2017-08-11 | textbox | 36.74% | 65.66% | 47.11% | |||
2015-03-31 | AJOU | 46.94% | 47.26% | 47.10% | |||
2015-03-30 | NJU_Text_Version1 | 38.32% | 56.33% | 45.62% | |||
2015-03-31 | NJU_Text_Version2 | 37.46% | 54.14% | 44.28% | |||
2017-10-12 | TextFCN V2 | 37.02% | 54.19% | 43.99% | |||
2015-04-02 | NJU_Text_Version5 | 37.84% | 51.41% | 43.59% | |||
2019-07-08 | psenet | 45.74% | 38.84% | 42.01% | |||
2015-04-02 | HUST MCLAB (VER3.0) | 37.79% | 44.00% | 40.66% | |||
2015-04-02 | HUST_MCLAB_VER1.0 | 34.81% | 47.47% | 40.17% | |||
2015-04-02 | HUST_MCLAB_VER2.0 | 34.09% | 46.49% | 39.33% | |||
2015-04-02 | HUST_MCLAB_VER.0 | 34.09% | 46.49% | 39.33% | |||
2015-04-02 | Deep2Text-MO | 32.11% | 49.59% | 38.98% | |||
2015-04-03 | CNN Proposal Based MSER | 34.42% | 34.71% | 34.57% | |||
2015-04-03 | TD-IMU | 25.28% | 34.56% | 29.20% | |||
2015-04-03 | TextCatcher-2 (LRDE) | 34.81% | 24.91% | 29.04% | |||
2018-07-11 | 111 | 21.71% | 35.21% | 26.86% | |||
2018-07-11 | 123 | 22.53% | 33.07% | 26.80% | |||
2018-07-11 | 2015_test_result_2000_2200_0.8 | 23.54% | 28.85% | 25.93% | |||
2018-04-25 | Full YOLO | 22.00% | 31.54% | 25.92% | |||
2018-07-08 | 123456 | 18.05% | 20.47% | 19.19% | |||
2018-04-25 | Tiny YOLO | 14.54% | 14.39% | 14.47% | |||
2015-04-01 | imagine | 5.44% | 12.97% | 7.67% |