Peicheng Wang

Pieter M. Blok, Haozhou Wang, Hyun Kwon Suh et al.

Potato yield is a key indicator for optimizing cultivation practices in agriculture. Potato yield can be estimated on harvesters using RGB-D cameras, which capture three-dimensional (3D) information of individual tubers moving along the conveyor belt. However, point clouds reconstructed from RGB-D images are incomplete due to self-occlusion, leading to systematic underestimation of tuber weight. To address this, we introduce PointRAFT, a high-throughput point cloud regression network that directly predicts continuous 3D shape properties, such as tuber weight, from partial point clouds. Rather than reconstructing full 3D geometry, PointRAFT infers target values directly from raw 3D data. Its key architectural novelty is an object height embedding that incorporates tuber height as an additional geometric cue, improving weight prediction under practical harvesting conditions. PointRAFT was trained and evaluated on 26,688 partial point clouds collected from 859 potato tubers across four cultivars and three growing seasons on an operational harvester in Japan. On a test set of 5,254 point clouds from 172 tubers, PointRAFT achieved a mean absolute error of 12.0 g and a root mean squared error of 17.2 g, substantially outperforming a linear regression baseline and a standard PointNet++ regression network. With an average inference time of 6.3 ms per point cloud, PointRAFT supports processing rates of up to 150 tubers per second, meeting the high-throughput requirements of commercial potato harvesters. Beyond potato weight estimation, PointRAFT provides a versatile regression network applicable to a wide range of 3D phenotyping and robotic perception tasks. The code, network weights, and a subset of the dataset are publicly available at https://github.com/pieterblok/pointraft.git.

Kejia Liu, Haoyang Zhou, Ruoyu Xu et al.

Recent advances in cross-view geo-localization (CVGL) methods have shown strong potential for supporting unmanned aerial vehicle (UAV) navigation in GNSS-denied environments. However, existing work predominantly focuses on matching UAV views to onboard map tiles, which introduces an inherent trade-off between accuracy and storage overhead, and overlooks the importance of the UAV's heading during navigation. Moreover, the substantial discrepancies and varying overlaps in cross-view scenarios have been insufficiently considered, limiting their generalization to real-world scenarios. In this paper, we present Bearing-UAV, a purely vision-driven cross-view navigation method that jointly predicts UAV absolute location and heading from neighboring features, enabling accurate, lightweight, and robust navigation in the wild. Our method leverages global and local structural features and explicitly encodes relative spatial relationships, making it robust to cross-view variations, misalignment, and feature-sparse conditions. We also present Bearing-UAV-90k, a multi-city benchmark for evaluating cross-view localization and navigation. Extensive experiments show encouraging results that Bearing-UAV yields lower localization error than previous matching/retrieval paradigm across diverse terrains. Our code and dataset will be made publicly available.

Pukun Zhao, Longxiang Wang, Chen Chen et al.

Structured spatial navigation is a core benchmark for Large Language Models (LLMs) spatial reasoning. Existing paradigms like Visualization-of-Thought (VoT) are prone to cascading errors in complex topologies. To solve this, we propose STAR, a two-stage framework grounded on topological anchors, and introduce the RedMaze-23K dataset with human-inspired turnpoint annotations. The first stage uses supervised fine-tuning to help models internalize spatial semantics and prune redundant paths. The second adopts Spatial-aware Segment-level Direct Preference Optimization (SDPO) to refine self-correction in long-horizon navigation. Experiments show STAR achieves state-of-the-art performance among open-source models: its 32B variant outperforms DeepSeek-V3 (29.27% vs. 25.00%) and reaches 82.4% of GPT-4's performance.

Yafei Wu, Jian Shi, Peicheng Wang et al.

As Android malware is growing and evolving, deep learning has been introduced into malware detection, resulting in great effectiveness. Recent work is considering hybrid models and multi-view learning. However, they use only simple features, limiting the accuracy of these approaches in practice. In this paper, we propose DeepCatra, a multi-view learning approach for Android malware detection, whose model consists of a bidirectional LSTM (BiLSTM) and a graph neural network (GNN) as subnets. The two subnets rely on features extracted from statically computed call traces leading to critical APIs derived from public vulnerabilities. For each Android app, DeepCatra first constructs its call graph and computes call traces reaching critical APIs. Then, temporal opcode features used by the BiLSTM subnet are extracted from the call traces, while flow graph features used by the GNN subnet are constructed from all the call traces and inter-component communications. We evaluate the effectiveness of DeepCatra by comparing it with several state-of-the-art detection approaches. Experimental results on over 18,000 real-world apps and prevalent malware show that DeepCatra achieves considerable improvement, e.g., 2.7% to 14.6% on F1-measure, which demonstrates the feasibility of DeepCatra in practice.

Cong Sun, Xinpeng Xu, Yafei Wu et al.

The misunderstanding and incorrect configurations of cryptographic primitives have exposed severe security vulnerabilities to attackers. Due to the pervasiveness and diversity of cryptographic misuses, a comprehensive and accurate understanding of how cryptographic misuses can undermine the security of an Android app is critical to the subsequent mitigation strategies but also challenging. Although various approaches have been proposed to detect cryptographic misuse in Android apps, studies have yet to focus on estimating the security risks of cryptographic misuse. To address this problem, we present an extensible framework for deciding the threat level of cryptographic misuse in Android apps. Firstly, we propose a general and unified specification for representing cryptographic misuses to make our framework extensible and develop adapters to unify the detection results of the state-of-the-art cryptographic misuse detectors, resulting in an adapter-based detection tool chain for a more comprehensive list of cryptographic misuses. Secondly, we employ a misuse-originating data-flow analysis to connect each cryptographic misuse to a set of data-flow sinks in an app, based on which we propose a quantitative data-flow-driven metric for assessing the overall risk of the app introduced by cryptographic misuses. To make the per-app assessment more useful for app vetting at the app-store level, we apply unsupervised learning to predict and classify the top risky threats to guide more efficient subsequent mitigation. In the experiments on an instantiated implementation of the framework, we evaluate the accuracy of our detection and the effect of data-flow-driven risk assessment of our framework. Our empirical study on over 40,000 apps and the analysis of popular apps reveal important security observations on the real threats of cryptographic misuse in Android apps.