Low-light object detection faces significant challenges due to the substantial domain shift between normal-light and low-light conditions. Prior works often enhance low-light images before detection, but this preprocessing can introduce artifacts that degrade detection performance since it focuses on human visual quality rather than task-specific features. Other methods incorporate illumination-aware modules for low-light feature learning, yet their scalability is limited by the scarcity of annotated low-light datasets. To overcome these limitations, we propose a unified Dual-Level Domain Adaptation (DLDA) framework that jointly addresses image-level and feature-level domain discrepancies. Specifically, we introduce a luminance-aware contrastive translation module that synthesizes target-style low-light images while preserving structural details, enabling effective image-level adaptation. Building on this, we further design a multi-scale conditional adversarial alignment strategy that promotes semantic consistency across feature hierarchies to enhance domain-invariant feature extraction. Extensive experiments on multiple low-light detection benchmarks demonstrate that DLDA achieves state-of-the-art performance, exhibiting strong robustness and generalization.

Xiangxi Zheng, Kuang He, Jiayi Hu, Ping Yu, Rui Yan, Yuan Yao, Peng Hou, Anxiang Zeng, Alex Jinpeng Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Third-person perspective (3PP) VR can afford dual embodiment, letting users act through an in-world avatar while simultaneously operating an omniscient entity (OE) from a detached viewpoint. We investigate how the sense of body ownership (SoBO) reallocates between these two asymmetric entities during continuous play. We systematically varied macro-level design factors—avatar and OE involvement—across three maze scenes and one combat scene in a within-subjects study. After each trial, participants produced time-aligned, continuous SoBO traces for both entities; gaze and controller signals were logged, and recall interviews captured first-person explanations. This research reveals three main findings. First, SoBO is not statically fixed but dynamically reallocated between the avatar and the OE across task phases. Second, allocation follows design factors: higher OE involvement in the maze strengthened ownership toward it, while combat consistently favored the avatar. Third, moment-to-moment gaze and control intensity predict ownership allocation, with fluency and salient events emerging as additional influences. These findings provide a framework for understanding dynamic SoBO and inform the design of adaptive 3PP VR systems.

EEG recordings obtained before medication are regarded as valuable biological indicators for depression detection. Currently, depression diagnosis based on EEG using convolutional neural networks (CNNs) has achieved relatively high detection performance, but some issues remain unresolved. CNNs are constrained by their limited receptive fields, which restrict them to capturing local rather than global dependencies. In addition, the complex features learned by CNNs are often hard to interpret and typically require a substantial number of trainable parameters. To tackle these issues, an interpretable hybrid neural network named SINCFORMER-SHAP is proposed. SINCFORMER-SHAP comprises two main components, namely the spatial-frequency and temporal feature extraction modules. The spatial-frequency feature extraction module leverages a hybrid design, where temporal filtering through a sinc-based convolution is coupled with spatial convolution, enabling the model to learn fine-grained spatial-spectral patterns. The sinc-convolutional layer helps constrain the parameter count, enhancing model efficiency. Subsequently, the temporal domain feature extraction module utilizes Transformer to capture global time-domain dependencies. Kernel visualization is used to provide direct insights into the spectral features learned by the spatial-frequency feature extraction module. To further enhance interpretability on the spatial domain, a post-hoc analysis is conducted using SHAP method. Based on the results of interpretability analysis, potential biomarkers have been observed within alpha and gamma rhythms across the frontal, parietal, temporal, and occipital areas. Comprehensive experiments conducted on public MODMA, EDRA and Mumtaz datasets were used to assess the performance of the proposed approach. The experimental outcomes provide compelling evidence that the proposed method not only surpasses multiple state-of-the-art approaches in performance, but also contributes a significant advancement toward the development of interpretable diagnostic technique for depression, thereby bridging the gap between computational methodologies and practical psychiatric applications.

The widespread deployment of control-flow integrity has shifted attackers' focus to non-control data attacks. In OS kernel exploits, attackers can gain root access or escalate privileges by corrupting critical non-control objects without hijacking the control flow. However, searching for exploitable non-control data in the OS kernel is challenging because of the data's semantic complexity and lack of universal patterns. This work represents the first study to semi-automatically discover and evaluate exploitable non-control data within the Linux kernel's file system, with minimal domain knowledge. Utilizing a custom analysis and testing framework, we identify promising candidate objects both statically and dynamically. We categorize these objects into types suitable for various exploit strategies, including a systematic strategy to overcome defenses that isolate many of these objects. These objects can be exploitable without requiring KASLR, thus making the exploits simpler and more reliable. We evaluate the exploitability of the file system objects using 18 real-world CVEs with various exploit strategies. We further develop 10 end-to-end exploits against the kernel with all state-of-the-art mitigations enabled.

Multimodal emotion recognition in conversation (MERC) aims to accurately capture speakers' emotional trajectories by integrating diverse sources such as text, visual cues, and speech. However, existing MERC methods still struggle with conflicts in cross-modal fusion and over-smoothing in deep graph networks. To address these challenges, we propose the Emotion-aware Graph Convolution-based Multimodal Integration Network (EmoMI), a multi-scale heterogeneous graph contrastive learning framework. EmoMI tackles these issues with two core innovations. First, to mitigate fusion conflicts, it employs a novel alternating aggregation scheme, systematically separating interand intra-modal information. Second, to combat over-smoothing, our multi-scale framework generates distinct local and global views of the dialogue graph. We then introduce prototype-level contrastive learning to align stable class “semantic anchors” across these views, providing a robust supervisory signal and enhancing feature discriminability. EmoMI establishes new state-of-the-art results on the IEMOCAP and MELD benchmarks, outperforming strong graph-based baselines. Our findings demonstrate that this synergistic approach effectively resolves long-standing challenges in MERC, paving the way for more robust conversational emotion analysis models.

Recently studies usually used remote sensing products and reanalysis datasets to obtain the precipitation. Yet, due to sparse ground observations and complex topography, the accuracy of precipitation products in mountainous regions is one of the great concerns in hydrology. Thus, assessing the accuracy and availability of precipitation products is vital to hydrological research in mountainous regions. In this study, we evaluated the precision of precipitation products including IMERG, ERA5-Land, and AERA5-Asia at multiple temporal and spatial scales using ground observations of meteorological stations from 2001 to 2015 in Southwest China, a typical mountainous region. The key findings are: (1) The IMERG datasets had the best performance on a monthly scale. (2) The IMERG and AERA5-Asia data products had higher precision at seasonal and annual scales. (3) The AERA5-Asia achieved better performance in the high- altitude areas compared to others. In addition, the ERA5- Land has the lowest accuracy. This study can provide important guidelines for the selection of precipitation products for both scientific and management purposes in mountainous areas, especially in Southwest China.

This paper analyzes the torque performance of the flat wire motor, a potential winding form of a high-performance motor. The motor has continuous flat wire(CFW), so it has no solder joint in every branch, which improves the reliability. Besides, the axis length of end windings is shorter than that of the hairpin motor, so its power density is higher. However, the torque ripple and the cogging torque of the CFW motor are higher because of the larger slot opening. This paper compares the rotor skew and the stator chute, reduces the torque ripple and the cogging torque by the rotor skew.