MXMingshuo Xu

Mingshuo Xu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Mingshuo Xu has a total of 6 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Beyond the Optic Lobe: Closing the Representational and Verification Gaps of STMD for Moving Target Proposal

    by Mingshuo Xu, Haonan Gong, Zhou Daniel Hao, Jigen Peng, Yi Zheng, Shigang Yue

    Bio-inspired small-target motion detectors (STMDs) provide pixel-level saliency but lack proposal geometry and moving-target verification. Inspired by insect visual pathways, we present Beyond the Optic Lobe (BotL), a moving-target proposal framework compatible with seven STMD front-ends. A learning-free ON/OFF-histogram bridge converts saliency into boxes, and a binary verifier scores moving targets versus clutter without predicting categories. Under strict matching, BotL raises the area under the Recall versus false positives per image (FPPI) curve (AUC) for the default vSTMD-F front-end from 0.607 to 0.836 on re-annotated RIST and from 0.109 to 0.611 on SeaDronesSee-Tiny-SAHI (STS). Strict AUC and candidate Recall at a 15-proposal budget improve on every RIST test sequence. STS gains are large for the vSTMD family, moderate for FracSTMD, small for HaarSTMD, and negligible for ESTMD, DSTMD, and STMD+. Stage analysis distinguishes coverage constraints from verifier ranking limitations. YOLO detectors are more accurate in in-domain comparisons, but BotL leads in STS-to-RIST zero-shot transfer, with AUC 0.252 versus 0.016 and 0.030. BotL uses 16,341 learned parameters and reduces floating-point operations by a factor of 21–41 relative to these two YOLO detectors, serving as a compute-efficient proposal stage.

  • vSTMD: Visual Motion Detection for Extremely Tiny Target at Various Velocities

    by Mingshuo Xu, Hao Luan, Zhou Daniel Hao, Jigen Peng, Shigang Yue

    Visual motion detection for extremely tiny (ET-) targets is challenging, due to their category-independent nature and the scarcity of visual cues, which often incapacitate mainstream feature-based models. Natural architectures with rich interpretability offer a promising alternative, where STMD architectures derived from insect visual STMD (Small Target Motion Detector) pathways have demonstrated their effectiveness. However, previous STMD models are constrained to a narrow velocity range, hindering their efficacy in real-world scenarios where targets exhibit diverse and unstable dynamics. To address this limitation, we present vSTMD, a learning-free model for motion detection of ET-targets at various velocities. vSTMD proposes two key mechanisms: cross-Inhibition Dynamic Potential (cIDP) and Collaborative Directional Gradient Calculation (CDGC). Specifically, cIDP serves as a self-adaptive mechanism, efficiently capturing motion cues across a wide velocity spectrum. CDGC enhances orienting accuracy and robustness while reducing computational overhead to one-eighth of previously isolated strategies. Evaluated on the real-world dataset RIST, the proposed vSTMD and its feedback-facilitated variant vSTMD-F achieve relative F1 gains of 30% and 58% over state-of-the-art STMD approaches, respectively. Furthermore, both models demonstrate competitive orientation estimation performance compared to SOTA deep learning-driven methods. Experiments also reveal the superiority of the natural architecture for ET-object motion detection - vSTMD is 60× faster than contemporary data-driven methods, making it highly suitable for real-time applications in dynamic scenarios and complex backgrounds. Code is available at Project Repository.