Code comments are essential for software development and maintenance, as they provide natural language descriptions of the code that help developers understand the program and reduce the time spent on comprehension. However, writing code comments can be tedious and time-consuming, and many software projects lack comprehensive and up-to-date comments, which can impair the readability and maintainability of programs.

Online Judge (OJ) system, which can automatically evaluate the results (right or wrong) of programs by executing them on standard test cases, is widely used in programming education. While an OJ system with personalized feedback can not only give execution results, but also provide information to assist students in locating their problems quickly. Automatically fault localization techniques are designed to find the exact faults in programs automatically, experimental results showed their effect on locating artificial faults, but their effectiveness on novice programs needs to be investigated. In this paper, we first evaluate the effectiveness of several widely-studied fault localization techniques on novice programs, and then we use fine-grained test cases to improve the fault localization accuracy. Empirical studies are conducted on 77 real student programs and the results show that, compared with original test cases in OJ system, the fault localization accuracy can be improved obviously when using fine-grained test cases. More specifically, in terms of TOP-1, TOP-3 and TOP-5 metrics, the maximum results can be improved from 5, 22, 37 to 9, 24, 48, respectively. The results indicate that more faults can be located when checking the top 1, 3 or 5 statements, so the fault localization accuracy is enhanced. Furthermore, a Test Case Granularity (TCG) concept is introduced to describe fine-grained test cases, and empirically studies demonstrate that there is a strong correlation between TCG and fault localization accuracy.

Scene text image contains two levels of contents: visual texture and semantic information. Although the previous scene text recognition methods have made great progress over the past few years, the research on mining semantic information to assist text recognition attracts less attention, only RNN-like structures are explored to implicitly model semantic information. However, we observe that RNN based methods have some obvious shortcomings, such as time-dependent decoding manner and one-way serial transmission of semantic context, which greatly limit the help of semantic information and the computation efficiency. To mitigate these limitations, we propose a novel end-to-end trainable framework named semantic reasoning network (SRN) for accurate scene text recognition, where a global semantic reasoning module (GSRM) is introduced to capture global semantic context through multi-way parallel transmission. The state-of-the-art results on 7 public benchmarks, including regular text, irregular text and non-Latin long text, verify the effectiveness and robustness of the proposed method. In addition, the speed of SRN has significant advantages over the RNN based methods, demonstrating its value in practical use.