Qingpeng Li

Qingpeng Li, Yuxin Zhang, Leyuan Fang et al.

Object detection algorithms are pivotal components of unmanned aerial vehicle (UAV) imaging systems, extensively employed in complex fields. However, images captured by high-mobility UAVs often suffer from motion blur cases, which significantly impedes the performance of advanced object detection algorithms. To address these challenges, we propose an innovative object detection algorithm specifically designed for blurry images, named DREB-Net (Dual-stream Restoration Embedding Blur-feature Fusion Network). First, DREB-Net addresses the particularities of blurry image object detection problem by incorporating a Blurry image Restoration Auxiliary Branch (BRAB) during the training phase. Second, it fuses the extracted shallow features via Multi-level Attention-Guided Feature Fusion (MAGFF) module, to extract richer features. Here, the MAGFF module comprises local attention modules and global attention modules, which assign different weights to the branches. Then, during the inference phase, the deep feature extraction of the BRAB can be removed to reduce computational complexity and improve detection speed. In loss function, a combined loss of MSE and SSIM is added to the BRAB to restore blurry images. Finally, DREB-Net introduces Fast Fourier Transform in the early stages of feature extraction, via a Learnable Frequency domain Amplitude Modulation Module (LFAMM), to adjust feature amplitude and enhance feature processing capability. Experimental results indicate that DREB-Net can still effectively perform object detection tasks under motion blur in captured images, showcasing excellent performance and broad application prospects. Our source code will be available at https://github.com/EEIC-Lab/DREB-Net.git.

Xin Yuan, Siqi Li, Jiateng Wei et al.

Pruning is an effective method for compressing Large Language Models, but finding an optimal, non-uniform layer-wise sparsity allocation remains a key challenge. While heuristic methods are fast but yield suboptimal performance, more powerful search-based approaches like Reinforcement Learning are often hindered by prohibitive computational costs on large-scale models. To overcome this efficiency barrier, we propose FastForward Pruning. Its core is a decoupled, single-step RL framework that separates policy optimization from the complex budget satisfaction problem. Such a decoupling is crucial for efficiently searching the vast policy space of LLMs. This curriculum-based strategy begins with low-cost, simple tasks and gradually increases in complexity, significantly reducing the search's computational overhead. Evaluated on the LLaMA, Mistral, and OPT model families, our framework discovers pruning policies that achieve superior performance over strong heuristic baselines. Crucially, when compared to other search-based algorithms, our method achieves competitive or superior results at a fraction of the computational cost, demonstrating a clear advantage in search efficiency.

Yuanbing Ouyang, Yizhuo Liang, Qingpeng Li et al.

Vision Transformers (ViTs) excel in semantic segmentation but demand significant computation, posing challenges for deployment on resource-constrained devices. Existing token pruning methods often overlook fundamental visual data characteristics. This study introduces 'LVTP', a progressive token pruning framework guided by multi-scale Tsallis entropy and low-level visual features with twice clustering. It integrates high-level semantics and basic visual attributes for precise segmentation. A novel dynamic scoring mechanism using multi-scale Tsallis entropy weighting overcomes limitations of traditional single-parameter entropy. The framework also incorporates low-level feature analysis to preserve critical edge information while optimizing computational cost. As a plug-and-play module, it requires no architectural changes or additional training. Evaluations across multiple datasets show 20%-45% computational reductions with negligible performance loss, outperforming existing methods in balancing cost and accuracy, especially in complex edge regions.

Qingpeng Li, Lichao Mou, Qizhi Xu et al.

Vehicle detection is a significant and challenging task in aerial remote sensing applications. Most existing methods detect vehicles with regular rectangle boxes and fail to offer the orientation of vehicles. However, the orientation information is crucial for several practical applications, such as the trajectory and motion estimation of vehicles. In this paper, we propose a novel deep network, called rotatable region-based residual network (R$^3$-Net), to detect multi-oriented vehicles in aerial images and videos. More specially, R$^3$-Net is utilized to generate rotatable rectangular target boxes in a half coordinate system. First, we use a rotatable region proposal network (R-RPN) to generate rotatable region of interests (R-RoIs) from feature maps produced by a deep convolutional neural network. Here, a proposed batch averaging rotatable anchor (BAR anchor) strategy is applied to initialize the shape of vehicle candidates. Next, we propose a rotatable detection network (R-DN) for the final classification and regression of the R-RoIs. In R-DN, a novel rotatable position sensitive pooling (R-PS pooling) is designed to keep the position and orientation information simultaneously while downsampling the feature maps of R-RoIs. In our model, R-RPN and R-DN can be trained jointly. We test our network on two open vehicle detection image datasets, namely DLR 3K Munich Dataset and VEDAI Dataset, demonstrating the high precision and robustness of our method. In addition, further experiments on aerial videos show the good generalization capability of the proposed method and its potential for vehicle tracking in aerial videos. The demo video is available at https://youtu.be/xCYD-tYudN0.