arXiv subjects

Shiyang Su

Publications and source records attributed to Shiyang Su.

3 recordsLinked to original sources

arXiv2026

Information Density Imbalance in Visual Object Detection

In object detection, the number of instances is typically used to determine whether a dataset exhibits a long-tailed distribution, implicitly assuming that the model will perform poorly on categories with fewer instances. This assumption has led to extensive research on category bias in datasets with imbalanced instance numbers. However, even in datasets where instance numbers are relatively balanced, models still exhibit category bias, indicating that instance count alone cannot explain this phenomenon. In this work, we first introduce the concept and measurement of information density. We then observe a significant negative correlation between a category's information density and its accuracy, and we investigate how the training process impacts this relationship. Empirical studies suggest that information density imbalance may be a potential source of category bias. To preliminarily validate the potential of information density, we made simple improvements to three advanced object detection loss functions using this concept. Experiments on the Pascal VOC, COCO-LT, and LVIS datasets demonstrate that information density can significantly reduce model bias while effectively enhancing the overall performance of existing loss functions. This study provides a new perspective for understanding the generalized bias phenomenon in object detection models and offers new tools for designing fairer loss functions and training strategies.

cs.CV↗

arXiv2026

MathDoc: Benchmarking Structured Extraction and Active Refusal on Noisy Mathematics Exam Papers

The automated extraction of structured questions from paper-based mathematics exams is fundamental to intelligent education, yet remains challenging in real-world settings due to severe visual noise. Existing benchmarks mainly focus on clean documents or generic layout analysis, overlooking both the structural integrity of mathematical problems and the ability of models to actively reject incomplete inputs. We introduce MathDoc, the first benchmark for document-level information extraction from authentic high school mathematics exam papers. MathDoc contains \textbf{3,609} carefully curated questions with real-world artifacts and explicitly includes unrecognizable samples to evaluate active refusal behavior. We propose a multi-dimensional evaluation framework covering stem accuracy, visual similarity, and refusal capability. Experiments on SOTA MLLMs, including Qwen3-VL and Gemini-2.5-Pro, show that although end-to-end models achieve strong extraction performance, they consistently fail to refuse illegible inputs, instead producing confident but invalid outputs. These results highlight a critical gap in current MLLMs and establish MathDoc as a benchmark for assessing model reliability under degraded document conditions. Our project repository is available at \href{https://github.com/winnk123/papers/tree/master}{GitHub repository}

cs.CV↗

arXiv2023

Reforming Physics Exams Using Openly Accessible Large Isomorphic Problem Banks created with the assistance of Generative AI: an Explorative Study

This paper explores using large isomorphic problem banks to overcome many challenges of traditional exams in large STEM classes, especially the threat of content sharing websites and generative AI to the security of exam items. We first introduce an efficient procedure for creating large numbers of isomorphic physics problems, assisted by the large language model GPT-3 and several other open-source tools. We then propose that if exam items are randomly drawn from large enough problem banks, then giving students open access to problem banks prior to the exam will not dramatically impact students' performance on the exam or lead to wide-spread rote-memorization of solutions. We tested this hypothesis on two mid-term physics exams, comparing students' performance on problems drawn from open isomorphic problem banks to similar transfer problems that were not accessible to students prior to the exam. We found that on both exams, both open bank and transfer problems had the highest difficulty. The differences in percent correct were between 5% to 10%, which is comparable to the differences between different isomorphic versions of the same problem type. Item response theory analysis found that both types of problem have high discrimination (>1.5) with no significant differences. Student performance on open-bank and transfer problems are highly correlated with each other, and the correlations are stronger than average correlations between problems on the exam. Exploratory factor analysis also found that open-bank and transfer problems load on the same factor, and even formed their own factor on the second exam. Those observations all suggest that giving students open access to large isomorphic problem banks only had a small impact on students' performance on the exam but could have significant potential in reforming traditional classroom exams.

physics.ed-ph↗