Deep-learning techniques have significantly improved object detection performance, especially with binocular images in 3D scenarios. To supervise the depth information in stereo 3D object detection, reconstructing the 3D dense depth of LiDAR point clouds causes higher computational costs and lower inference speed. After exploring the intrinsic relationship between the implicit depth information and semantic texture features of the binocular images, we propose an efficient and accurate 3D object detection algorithm, FCNet, in stereo images. First, we construct a multi-scale cost–volume containing implicit depth information using the normalized dot-product by generating multi-scale feature maps from the input stereo images. Secondly, the variant attention model enhances its global and local description, and the sparse region monitors the depth loss deep regression. Thirdly, for balancing the channel information preservation of the re-fused left–right feature maps and computational burden, a reweighting strategy is employed to enhance the feature correlation in merging the last-layer features of binocular images. Extensive experiment results on the challenging KITTI benchmark demonstrate that the proposed algorithm achieves better performance, including a lower computational cost and higher inference speed in 3D object detection.
Remote sensing image classification plays a crucial role in scene understanding task interpretation. Currently, remote sensing image classification is constrained by the complexity, diversity, and time-varying nature of the scene as well as the variability of spatio-temporal resolution while the classification detection accuracy is difficult to be further improved. In this paper, a new hybrid cross-activation network (HC-Mamba) is proposed based on the Visual State Space (VSS) model represented by Mamba to further improve the understanding of image collation and provide more quasi-clustered semantic cues for scene recognition. Firstly, we further improve the classification performance by combining the features of hardware-aware design, global receptive domain and linear modeling complexity. Secondly, we propose a local content enrichment module to learn to distinguish local features to maintain the essential characteristics of image sequences. Finally, we propose a dynamic hybrid cross-activation mechanism to counteract the limitations of causal sequence modeling that makes it difficult to adapt to 2D image data, and further improve the model's ability to model non-causal data. In addition, HC-Mamba maintains the inherent modeling mechanism of the original model, and demonstrates better performance on three benchmark remote sensing image classification data through a large number of experiments, which indicates that our proposed model has a great potential for the backbone network in future visualization-based modeling work.
Salient object detection in optical remote sensing images (RSI-SOD) is a very challenging task due to the complex scale, shape, details and uncertainty of the predicted location of the object, etc. In this paper, we propose a novel important pick-and-pass gate refinement network model for SOD in optical RSIs, named IP2GRNet, including a detail refinement stage and a feature pick-and-pass refinement stage. Specifically, the detail refinement stage, named refinement synchronization, uses the self-modality attention refinement (SAR) module and the dynamic weight refinement (DWR) module to accurately describe and capture approximate coordinate positions and feature information of salient objects. The feature pick-and-pass and refinement stage progressively picks and refines the prediction results in a coarse to fine manner by combining high-level semantic information and low-level semantic information under the guidance of attention and counter-attention. In addition, the aggregation operation is used to fuse detail refinement information as well as mixture loss for supervised network training, which effectively improves the model performance from three perspectives: pixel, region, and statistics. Extensive experiments on two benchmarks datasets demonstrate that our proposed IP2GRNet on both qualitatively and quantitatively is outperforms to the state-of-the-art saliency detectors.