极市导读
最近,Hamming OCR提出了Hamming Embedding和LSH局部敏感哈希分类算法。它能在保留模型能力的同时,大幅度削减场景文本识别中所需模型的大小。>>加入极市CV技术交流群,走在计算机视觉的最前沿
Hamming OCR 是一个基于Transformer注意力的超轻量级文本识别模型,主要基于LSH局部敏感哈希编码和Max-Margin Loss的学习算法。
Hamming OCR: A Locality Sensitive Hashing Neural Networkfor SceneText Recognition
场景文本识别中很多模型都使用了笨重的模型,这些模型很难在移动端设备上部署。这也是最近大火的Paddle OCR使用CRNN实现其超轻量级识别模型的原因。PaddleOCR采用的策略就是暴力削减特征通道来减小模型,但是这样使得性能大幅度下降。最近在arXiv上公开的Hamming OCR提出了Hamming Embedding和LSH局部敏感哈希分类的算法,大幅度削减模型大小,同时保留模型能力。
我们首先看FC+softmax分类层:
参考文献
1. Lu, N.; Yu,W.; Qi, X.; Chen, Y.; Gong, P.; and Xiao, R. 2019. Master: Multi-aspectnon-local network for scene text recognition. arXiv preprint arXiv:1910.02562.
2. Li, H.;Wang, P.; Shen, C.; and Zhang, G. 2019. Show, attend and read: A simple andstrong baseline for irregular text recognition. In Proceedings of the AAAIConference on Artificial Intelligence, volume 33, 8610–8617.
3. Shi, B.;Bai, X.; and Yao, C. 2016. An end-to-end trainable neural network forimage-based sequence recognition and its application to scene text recognition.IEEE transactions on pattern analysis and machine intelligence 39(11):2298–2304.
4. Lan, Z.;Chen, M.; Goodman, S.; Gimpel, K.; Sharma, P.; and Soricut, R. 2019. Albert: Alite bert for selfsupervised learning of language representations. arXiv preprintarXiv:1909.11942 .
推荐阅读