Accurate segmentation of line-like structures in biomedical images (e.g., vasculature, muscle fibers) is critical, for their morphology, density, and connectivity serve as vital biomarkers for analysis, yet it remains challenging. Mainstream segmentation models have massive scale and produce fragment filamentous structures, while specialized methods often require complex post-processing. Therefore, we propose Line-like Structure Segmentation Network (LSSeg), an ultra-lightweight hierarchical architecture (28 K parameters) that directly outputs predictions for this specific task. LSSeg uniquely integrates segmentation and edge detection strengths by combining multi-level feature perception (via Feature Decimation Modules) with locally aware reconstruction (via Focus Locally Block) to achieve high-quality output while maintaining extreme efficiency (2.1 GFLOPs). Experiments across diverse datasets (retinal vessels, muscle aponeurosis, microtubule) show LSSeg outperforms universal segmentation models as well as edge detection models. The code is available at https://github.com/BM-AI-Lab/LSSeg.

Directed Greybox Fuzzing (DGF) is a technique used to efficiently cover specific program locations, making it suitable for vulnerability discovery and regression testing. However, previous DGF tools often suffer from the issue of getting trapped in local optima, where some targets are adequately covered while other targets remain unexploited. We propose the concept of Target-Directed Basic Blocks to efficiently guide the fuzzer, a multi-dimensional energy scheduling algorithm to guide the generation of seeds, and an adaptive exploration-exploitation strategy to prevent falling into local optima. Moreover, we optimize the mutation algorithm by employing a branch-sensitive strategy and a reinforcement learning algorithm based on the multi-armed bandit model. We implemented these algorithms in a tool called SAFuzz and conducted extensive experiments and evaluations in real-world scenarios. The results demonstrate that SAFuzz is effective in bug reproduction, guiding different targets, and validating static analysis results compared to state-of-the-art benchmark tools, i.e., AFLGo, Windranger, AFL++, and Lolly. Moreover, SA Fuzz has detected seven new vulnerabilities in real-world programs, and one of them could not be found by other baseline fuzzers within the time budget.

Directed greybox fuzzing (DGF) can quickly discover or reproduce bugs in programs by seeking to reach a program location or explore some locations in order. However, due to their static stage division and coarse-grained energy scheduling, prior DGF tools perform poorly when facing multiple target locations (targets for short). In this paper, we present multiple targets directed greybox fuzzing which aims to reach multiple programs locations in a fuzzing campaign. Specifically, we propose a novel strategy to adaptively coordinate exploration and exploitation stages, and a novel energy scheduling strategy by considering more relations between seeds and target locations. We implement our approaches in a tool called LeoFuzz and evaluate it on crash reproduction, true positives verification, and vulnerability exposure in real-world programs. Experimental results show that LeoFuzz outperforms six state-of-the-art fuzzers, i.e., QYSM, AFLGo, Lolly, Berry, Beacon and WindRanger in terms of effectiveness and efficiency. Moreover, LeoFuzz has detected 23 new vulnerabilities in real-world programs, and 12 of them have been assigned CVE IDs.