文本识别(效果汇总篇),持续更新
共 1878字,需浏览 4分钟
·
2022-02-09 17:36
最近看了几篇 近几年的 经典的 结果靠前的 文本识别相关的论文 现在 把他们总结下来
看的论文 各有各的优点 做实验时 往往需要选择一个好的backbone 但是性能始终记不住
下面记录看过的文章的识别准确率 方便针对不同需要的数据集 找到合适的backbone
下图是链接 以及部分见解 欢迎指正 持续更新
雨落棱角:RNN 与文本识别(系列一)RARE(CVPR): Robust Scene Text Recognition with Automatic Rectification
ASTER(TPAMI): https://ieeexplore.ieee.org/iel7/34/4359286/08395027.pdf
RARE和ASTER 同根同源 很相似
雨落棱角:RNN与文本识别(系列三)注意力机制Baek et al.(ICCV): What Is Wrong With Scene Text Recognition Model Comparisons? Dataset and Model Analysis
雨落棱角:RNN与文本识别(系列二)CRNN哪个更好?DAN(AAAI): Decoupled Attention Network for Text Recognition
雨落棱角:RNN与文本识别(系列四)--无需依靠序列的注意力机制(DAN)SAR(AAAI): Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition
雨落棱角:RNN与文本识别(系列五)--2D注意力机制ScRN(ICCV): Symmetry-constrained Rectification Network for Scene Text Recognition
MORAN(Pattern Recognition): A Multi-Object Rectified Attention Network for Scene Text Recognition
雨落棱角:RNN与文本识别(系列六)--修正网络加改正的注意力SE-ASTER(CVPR): SEED: Semantics Enhanced Encoder-Decoder Framework for Scene Text Recognition
SE_SAR(CVPR): SEED: Semantics Enhanced Encoder-Decoder Framework for Scene Text Recognition
SE_SAR是SE-ASTER这篇文章的补充方法
雨落棱角:RNN与文本识别(系列七)--语义增强的注意力机制TextScanner(AAAI): TextScanner: Reading Characters in Order for Robust Scene Text Recognition
SRN(CVPR): Towards Accurate Scene Text Recognition with Semantic Reasoning Networks
CSTR(arxiv): Revisiting Classification Perspective on Scene Text Recognition
雨落棱角:基于CNN的文本识别(可与序列识别比拟)SATRN(CVPR): On Recognizing Texts of Arbitrary Shapes with 2D Self-Attention
雨落棱角:Transformer与文本识别(系列一)Hamming OCR(arxiv): Hamming OCR: A Locality Sensitive Hashing Neural Network for Scene Text Recognition
RobustScanner(ECCV): http://link.springer.com/chapter/10.1007/978-3-030-58529-7_9
恰饭的坤:RNN与文本识别(系列八)--注意力机制方法的深度思考