Recent advancements in large language models (LLMs) have showcased significant improvements in mathematics. However, traditional math benchmarks like GSM8k offer a unidimensional perspective, falling short in providing a holistic assessment of the LLMs' math capabilities. To address this gap, we introduce MathBench, a new benchmark that rigorously assesses the mathematical capabilities of large language models. MathBench spans a wide range of mathematical disciplines, offering a detailed evaluation of both theoretical understanding and practical problem-solving skills. The benchmark progresses through five distinct stages, from basic arithmetic to college mathematics, and is structured to evaluate models at various depths of knowledge. Each stage includes theoretical questions and application problems, allowing us to measure a model's mathematical proficiency and its ability to apply concepts in practical scenarios. MathBench aims to enhance the evaluation of LLMs' mathematical abilities, providing a nuanced view of their knowledge understanding levels and problem solving skills in a bilingual context. The project is released at https://github.com/open-compass/MathBench .
Vascular diseases have long been regarded as a significant health concern. Accurately detecting the location, shape, and afflicted regions of blood vessels from a diverse range of medical images has proven to be a major challenge. Obtaining blood vessels that retain their correct topological structures is currently a crucial research issue. Numerous efforts have sought to reinforce neural networks' learning of vascular geometric features, including measures to ensure the correct topological structure of the segmentation result's vessel centerline. Typically, these methods extract topological features from the network's segmentation result and then apply regular constraints to reinforce the accuracy of critical components and the overall topological structure. However, as blood vessels are three-dimensional structures, it is essential to achieve complete local vessel segmentation, which necessitates enhancing the segmentation of vessel boundaries. Furthermore, current methods are limited to handling 2D blood vessel fragmentation cases. Our proposed boundary attention module directly extracts boundary voxels from the network's segmentation result. Additionally, we have established an optimal connection model based on minimal surfaces to determine the connection order between blood vessels. Our method achieves state-of-the-art performance in 3D multi-class vascular segmentation tasks, as evidenced by the high values of Dice Similarity Coefficient (DSC) and Normalized Surface Dice (NSD) metrics. Furthermore, our approach improves the Betti error, LR error, and BR error indicators of vessel richness and structural integrity by more than 10% compared to other methods, and effectively addresses vessel fragmentation and yields blood vessels with a more precise topological structure.