Drone-based vehicle detection aims at detecting vehicle locations and categories in aerial images. It empowers smart city traffic management and disaster relief. Researchers have made a great deal of effort in this area and achieved considerable progress. However, because of the paucity of data under extreme conditions, drone-based vehicle detection remains a challenge when objects are difficult to distinguish, particularly in low-light conditions. To fill this gap, we constructed a large-scale drone-based RGB-infrared vehicle detection dataset called DroneVehicle, which contains 28, 439 RGB-infrared image pairs covering urban roads, residential areas, parking lots, and other scenarios from day to night. Cross-modal images provide complementary information for vehicle detection, but also introduce redundant information. To handle this dilemma, we further propose an uncertainty-aware cross-modality vehicle detection (UA-CMDet) framework to improve detection performance in complex environments. Specifically, we design an uncertainty-aware module using cross-modal intersection over union and illumination estimation to quantify the uncertainty of each object. Our method takes uncertainty as a weight to boost model learning more effectively while reducing bias caused by high-uncertainty objects. For more robust cross-modal integration, we further perform illumination-aware non-maximum suppression during inference. Extensive experiments on our DroneVehicle and two challenging RGB-infrared object detection datasets demonstrated the advanced flexibility and superior performance of UA-CMDet over competing methods. Our code and DroneVehicle will be available: https://github.com/VisDrone/DroneVehicle .

Drones equipped with cameras (UAVs) can dynamically track the target in the air from a broader view compared with static cameras or moving sensors over the ground. However, it is still challenging to accurately track the target using a single drone due to several factors such as appearance variations and severe occlusions. To this end, we collect a new Multi-Drone single Object Tracking (MDOT) dataset that consists of 92 groups of video clips with 113,918 high resolution frames taken by two drones and 63 groups of video clips with 145,875 high resolution frames taken by three drones. Besides, two evaluation metrics are specially designed for multi-drone single object tracking, i.e., automatic fusion score (AFS) and ideal fusion score (IFS). Moreover, the agent sharing network (ASNet) is proposed by integrating self-supervised template sharing, target re-detection, and view-aware fusion of the target from multiple drones into a unified framework, which can improve the tracking accuracy significantly compared with single drone tracking. Extensive experiments on MDOT show that our ASNet significantly outperforms recent state-of-the-art trackers. The dataset can be found in https://github.com/VisDrone/MultiDrone.

Intelligent retail containers (IRCs) are a representative application of object detection in open-world scenarios. However, their wide application is limited because of labor-intensive data labeling and frequently updated retail goods categories. To address the two challenges, we propose a self-supervised fully automatic learning machine (SFLM), which consists of a self-paced active learning (SPAL) human–machine cooperation module and a clustering autolearning (CAL) module. Based on active learning and self-paced learning, we use the SPAL module to handle the labor-intensive data labeling task, which speeds up data labeling and reduces its cost. We further use the CAL module to overcome the challenge of frequently updated goods categories in intelligent retail. Considering the lack of large-scale datasets in real retail container scenarios, we propose a new challenging IRC dataset. This dataset consists of 35000 images, with 370000 bounding boxes and instance segmentation masks, covering 200 common categories of goods for different data acquisition settings. Extensive experimental results on the proposed dataset show that the proposed SFLM greatly reduced the labeling cost and time consumption and achieved superior performance in recognizing new categories. The dataset will be available at https://github.com/SunYM2020/IRC .

This paper has proposed a new baseline deep learning model of more benefits for image classification. Different from the convolutional neural network(CNN) practice where filters are trained by back propagation to represent different patterns of an image, we are inspired by a method called "PCANet" in "PCANet: A Simple Deep Learning Baseline for Image Classification?" to choose filter vectors from basis vectors in frequency domain like Fourier coefficients or wavelets without back propagation. Researchers have demonstrated that those basis in frequency domain can usually provide physical insights, which adds to the interpretability of the model by analyzing the frequencies selected. Besides, the training process will also be more time efficient, mathematically clear and interpretable compared with the "black-box" training process of CNN.

Visible-infrared (RGB-IR) collaborative perception is essential for robust multispectral object detection under challenging illumination and aerial scenarios.However, RGB and IR modalities inherently exhibit asymmetric reliability and information density across different physical environments. Although existing methods alleviate cross-modal inconsistency through attention mechanisms or feature alignment strategies, many of them still lack explicit reliability-conditioned role assignment, making it difficult to adaptively handle asymmetric information across modalities.To address these limitations, we propose the Asymmetric Guided Mamba-based Mixture-of-Experts Network (AGM$^2$Net), a context-aware expert scheduling framework that reformulates RGB-IR fusion as reliability-aware asymmetric information allocation, where learned routing distributions provide an interpretable proxy for modality-dependent expert preferences.For intra-modal processing, we propose the Adaptive Wavelet Enhanced Downsampling (AWEDown) module, which performs asymmetric frequency purification by selectively suppressing redundant low-frequency structural responses while preserving informative high-frequency details.For inter-modal interaction, we design the MambaMoE module, where heterogeneous experts and sparse routing dynamically select spatial, global, and directional query-modulate interaction policies for reliability-aware cross-modal refinement.Furthermore, to alleviate semantic-spatial inconsistency during multi-scale integration, we introduce the Asymmetric Semantic-Guided Fusion (ASGFusion) module, where semantically rich deep features selectively guide the filtering and refinement of shallow spatial details.Experiments on LLVIP, FLIR-aligned, and DroneVehicle show that AGM$^2$Net achieves state-of-the-art results, with 97.6\%, 82.6\%, and 85.2\% $mAP_{0.5}$, respectively.The source code is available at \href{https://github.com/HZAI-ZJNU/AGM2Net}{https://github.com/HZAI-ZJNU/AGM2Net}.