Object detection in unmanned aerial vehicle (UAV) imagery presents unique challenges, including extreme scale variations, highly imbalanced spatial distribution of objects, and computational constraints inherent to drone platforms. To address these issues, this paper introduces a novel self-regularized object detection framework specifically designed for UAV perspectives. The proposed method adopts LSNet as the backbone network, guided by the "See Large, Focus Small" principle, to enhance feature extraction capability for objects of varying scales under wide-angle UAV views. The framework incorporates two key components: a Self-Regularized Temporal Ratio Encoder (SRTR Encoder) that stabilizes training and improves bounding box quality by leveraging temporal aspect ratio constraints, and an Entropy Constrained Sampler (EC Sampler) that selects informative hard negative samples based on information entropy, thereby suppressing less relevant background regions and increasing training efficiency. Extensive experiments conducted on the VisDrone benchmark demonstrate that the proposed approach significantly outperforms existing methods, achieving a notable improvement in mean Average Precision (mAP) while simultaneously reducing computational costs. For instance, when integrated with the CEASC detector, the framework increases mAP from 25.4% to 28.0% while reducing GFLOPs from 105.90 to 92.61. The results validate the effectiveness of our method in addressing the specific challenges of UAV-based object detection, offering a balanced solution for accuracy and efficiency in resource-constrained scenarios.
Chenguang Zhang, Yangming Guo, Jian-Long Yu et al.· International Conference on...· 0 citations
In Class-Imbalanced Semi-Supervised Learning (CISSL), classifiers suffer from severe confirmation bias, particularly when the class distribution of unlabeled set are unknown or mismatched with that of the labeled set. Existing debiasing methods often rely on impractical assumptions for classifier bias estimation, such as known class priors or risky external proxies. To overcome these limitations, we propose Intrinsic Logit-Based Debiasing (ILBD), a robust post-hoc framework that estimates bias directly from task-relevant data without external dependencies and class priors. ILBD constructs a comprehensive bias estimator by decomposing output logits into two distinct components: (1) Label-Free Intrinsic Bias, derived from non-target K-1 logits by masking target-class maximum logits to capture the model's background distributional skew; and (2) Label-Guided Learning Bias, derived from target-class maximum logits to quantify confidence disparities between head and tail classes arising from varying learning difficulties. To ensure estimation reliability, we further employ a class-adaptive threshold to filter noisy pseudo-labels. By subtracting this estimated bias at inference, ILBD effectively rectifies the classifier's bias directly from the inherent statistical patterns of the training data, without the need for target-prior assumptions or external data. Extensive experiments on CIFAR-10/100-LT, STL-10-LT, and Small-ImageNet-127 demonstrate the effectiveness of ILBD. Our code is available at https://github.com/aroid721/ILBD.
