NLPIR SEMINAR 18th ISSUE COMPLETED

Last Monday, Gang Wang gave a presentation about the paper, End-to-End Text Recognition with Convolutional Neural Networks, and shared some opinion on it.

The paper is published on ICPR in 2012. The experiments were taken on two dataset: 1) ICDAR(International Conference on Document Analysis and Recognition ) 2003 Dataset and 2) SVT(Street View Text) Dataset. So their results were mainly concerned with English.

One question “In result table2, why I-5 is more accurate than I-50?” was asked. A possible answer is that: 5 and 50 are the number of distractor words provided by other research. The greater the numerical value, the louder the noise.