VALSE 论文速览 第161期：用于场景文本识别的自监督字符到字符蒸馏网络 ...

论文题目：

Self-Supervised Character-to-Character Distillation for Text Recognition

作者列表：

官同坤 (上海交通大学)、沈为 (上海交通大学)、杨学 (上海交通大学)、冯琦 (上海交通大学)、姜泽坤 (上海交通大学)、杨小康 (上海交通大学)

B站观看网址：

https://www.bilibili.com/video/BV1zQ4y1j7YT/

论文摘要：

After many years of supervised text recognition research, one question in the community is how to explore the potential of unlabeled real images by self-supervised learning. This talk will present our recent work in self-supervised text recognition from a different perspective: taking characters rather than sequences as the basic items for representation learning. We will also discuss how to keep their feature consistency learning under flexible augmentations.

参考文献：

[1] Tongkun Guan, Wei Shen, Xue Yang, Qi Feng, Zekun Jiang, Xiaokang Yang, “Self-Supervised Character-to-Character Distillation for Text Recognition,” in Proceeding of IEEE International Conference on Computer Vision (ICCV 2023), Paris, France, October 2023.

论文链接：

[https://openaccess.thecvf.com/content/ICCV2023/papers/Guan_Self-Supervised_Character-to-Character_Distillation_for_Text_Recognition_ICCV_2023_paper.pdf]

代码链接：

[https://github.com/TongkunGuan/CCD]

视频讲者简介：

Tongkun Guan is a PhD student at the Artificial Intelligence Institute, Shanghai Jiao Tong University. He received the M.S. degree in the Department of Automation from Shanghai Jiao Tong University, Shanghai, China, in 2023, and received the B.S. degree in Electrical Engineering and Automation from Hunan University, Changsha, China, in 2020. His research interests mainly include computer vision, text detection, and text recognition.

个人主页：

https://tongkunguan.github.io/

特别鸣谢本次论文速览主要组织者：

月度轮值AC：李爽 (北京理工大学)