To overcome the limitations of inadequate local feature representation and the underutilization of global information in dynamic graph convolutions, we propose a network that combines attention mechanisms with dual graph convolutions. Firstly, we construct a static graph based on the dynamic graph using the K-nearest neighbors algorithm and geometric distances of point clouds. This integration of dynamic and static graphs forms a dual graph structure, compensating for the underutilization of geometric positional relationships in the dynamic graph. Next, edge convolutions are applied to extract edge features from the dual graph structure. To further enhance the capturing ability of local features, we employ attention pooling, which combines max pooling and average pooling operations. Secondly, we introduce channel attention modules and spatial self-attention modules to improve the representation ability of global features and enhance semantic segmentation accuracy in our network. Experimental results on the S3DIS dataset demonstrate that compared to dynamic graph convolution alone, our proposed approach effectively utilizes both semantic and geometric relationships between point clouds using dual graph convolutions while addressing limitations related to insufficient local feature extraction. The introduction of attention mechanisms helps mitigate underutilization issues with global information, resulting in significant improvements in model performance.

Due to the huge difference in the representation of sketches and 3D models, sketch-based 3D model retrieval is a challenging problem in the areas of graphics and computer vision. Some state-of-the-art approaches usually extract features from 2D sketches and produce multiple projection views of 3D models, and then select one view of 3D models to match sketch. It’s hard to find “the best view” and views from different perspectives of a 3D model may be completely different. Other methods apply learning features to retrieve 3D models based on 2D sketch. However, sketches are abstract images and are usually drawn subjectively. It is difficult to be learned accurately. To address these problems, we propose cross-domain correspondence method for sketch-based 3D model retrieval based on manifold ranking. Specifically, we first extract learning features of sketches and 3D models by two-parts CNN structures. Subsequently, we generate cross-domain undirected graphs using learning features and semantic labels to create correspondence between sketches and 3D models. Finally, the retrieval results are computed by manifold ranking. Experimental results on SHREC 13 and SHREC 14 datasets show the superior performance in all 7 standard metrics, compared to the state-of-the-art approaches.

Sketch-based cross-domain visual data retrieval is the process of searching for images or 3D models using sketches as input. Achieving feature alignment is a significantly challenging task due to the high heterogeneity of cross-domain data. However, the alignment process faces significant challenges, such as domain gap, semantic gap, and knowledge gap. The existing methods adopt different ideas for sketch-based image and 3D shape retrieval tasks, one is domain alignment, and the other is semantic alignment. Technically, both tasks verify the accuracy of extracted features. Hence, we propose a method based on the global feature correlation and the feature similarity for multiple sketch-based cross-domain retrieval tasks. Specifically, the data from various modalities are fed into separate feature extractors to generate original features. Then, these features are projected to the shared subspace. Finally, domain consistency learning, semantic consistency learning, feature correlation learning and feature similarity learning are performed jointly to make the projected features modality-invariance. We evaluate our method on multiple benchmark datasets. Where the MAP in Sketchy, TU-Berlin, SHREC 2013 and SHREC 2014 are 0.466, 0.473, 0.860 and 0.816. The extensive experimental results demonstrate the superiority and generalization of the proposed method, compared to the state-of-the-art approaches. The in-depth analyses of various design choices are also provided to gain insight into the effectiveness of the proposed method. The outcomes of this research contribute to advancing the field of sketch-based cross-domain visual data retrieval and are expected to be applied to a variety of applications that require efficient retrieval of cross-domain domain data.

Compared with traditional artificial neural networks (ANNs), spiking neural networks (SNNs) have lower computational complexity, lower energy consumption, and faster inference speed, making them more promising for practical deployment on edge devices. However, in SNNs, since the output spikes of neurons are discrete, the model may face information loss, especially when the membrane potential is quantized into binary spikes, where quantization errors can lead to model precision loss and information loss. To address these challenges, this study proposes a spiking neural network based on hierarchical residual quantization and temporal error compensation (HRQ-TEC-SNN) and used for remote sensing object detection tasks. Through hierarchical residual quantization and temporal error compensation design, higher resolution quantization of membrane potential can be performed and the quantization threshold of membrane potential can be dynamically adjusted to compensate for errors introduced during the quantization process, allowing for fine-tuning at each time step and reducing information loss caused by coarse quantization. In terms of network structure, by introducing depthwise separable convolution modules, channel attention and spatial attention mechanisms, and integrating fast spatial pyramid pooling based on pulse neural networks, the model’s detection accuracy has been further improved while reducing the number of model parameters and computational costs. Experimental results show that the HRQ-TEC-SNN achieves significant advantages in both accuracy and energy consumption on the DOTAv1.0, DOTAv1.5 and DIOR datasets.