Jinyan Chen

Fan Yuan, Yuchen Yan, Yifan Jiang et al.

Vision language models (VLMs) achieve unified modeling of images and text, enabling them to accomplish complex real-world tasks through perception, planning, and reasoning. Among these tasks, reasoning is particularly representative, with mathematical reasoning serving as a prominent example. It highlights the high-level capability of VLMs to comprehend mathematical information in images and to perform sophisticated reasoning. Recently, numerous visual mathematical reasoning benchmarks have been proposed, but they are often restricted to geometry, lack coverage of math word problems, and rarely assess reasoning across multiple images. To address these gaps, we introduce GSM8K-V, a purely visual multi-image mathematical reasoning benchmark. GSM8K-V is built by systematically mapping each sample from the widely used text-based GSM8K into visual form. Through a carefully designed automated image-generation pipeline combined with meticulous human annotation, we curate 1,319 high-quality samples. We evaluate a wide range of open-source and closed-source models on GSM8K-V. Results show that although existing VLMs have nearly saturated performance on text-based GSM8K, there remains substantial room for improvement on GSM8K-V. For example, the best-performing model, Gemini-2.5-Pro, achieves 95.22% accuracy on GSM8K but only 46.93% on GSM8K-V. We conduct a comprehensive analysis of GSM8K-V, examining the limitations of current models as well as potential directions for improvement. GSM8K-V offers a new perspective on visual mathematical reasoning and establishes a benchmark to guide the development of more robust and generalizable VLMs.

Yucheng Jin, Jinyan Chen, Ziyue He et al.

Human pose estimation in low-resolution videos presents a fundamental challenge in computer vision. Conventional methods either assume high-quality inputs or employ computationally expensive cascaded processing, which limits their deployment in resource-constrained environments. We propose STAR-Pose, a spatial-temporal adaptive super-resolution framework specifically designed for video-based human pose estimation. Our method features a novel spatial-temporal Transformer with LeakyReLU-modified linear attention, which efficiently captures long-range temporal dependencies. Moreover, it is complemented by an adaptive fusion module that integrates parallel CNN branch for local texture enhancement. We also design a pose-aware compound loss to achieve task-oriented super-resolution. This loss guides the network to reconstruct structural features that are most beneficial for keypoint localization, rather than optimizing purely for visual quality. Extensive experiments on several mainstream video HPE datasets demonstrate that STAR-Pose outperforms existing approaches. It achieves up to 5.2% mAP improvement under extremely low-resolution (64x48) conditions while delivering 2.8x to 4.4x faster inference than cascaded approaches.