Search

Deep learning has shown remarkable performance in quantity of vision tasks. However, its large network generally requires quantity of samples to support sufficient parameters learning during training process. Such high request greatly reduces efficiency when applying on a small dataset with few samples. To alleviate this problem, we propose a novel data enhancement method for few-shot learning via a cutout approach and feature enhancement. After enhancement, the generated network not only produces distinguish feature map without collecting more samples, but also achieves advantage of feature representation with high efficiency for computing. Specifically, cutout approach is simple yet highly effective for image regulation, which enhances input image matrix by adding a fixed mask to improve robustness and overall performance of network. Afterward, we perform feature enhancement by proposing a feature promotion module, which uses characteristics of dilated convolution and sequential processing to improve feature representation ability, thus improving efficiency of the whole network. We conduct comparative experiments on both miniImageNet and CUB datasets, where the proposed method is superior to comparative methods in both 1-shot and 5-shot cases. [ABSTRACT FROM AUTHOR]

Point cloud representation is a challenge to extracting sufficient semantic information while ensuring that the sparsely point cloud spatial structure is complete. Benefiting from the Transformer network, recent studies have promoted the development of point cloud representation by extracting refined attention features based on global context. However, there is still undesired semantic information loss in the feature extraction stage. Hence, this paper proposes a novel architecture for 3D point cloud representation, namely Relation-Shape Transformer Network (RS-TNet), to address above problem while maintaining the merits of relation-shape embedding mechanism so as to generate rich and robust local semantic features. Specifically, RS-TNet can achieve coarse-to-fine grained semantic information coverage by integrating the global multi-head self-attention and local Relation-Feature extraction module simultaneously. Moreover, theoretical analysis demonstrates that RS-TNet can explicitly introduce the spatial relation of points by learning underlying shapes. In this way, extracted features are of more shape awareness and robustness. As a result, the proposed RS-TNet achieves 90.9% class accuracy and 85.6% Intersection-over-Union on ModelNet40 and ShapeNet datasets, respectively. Further, ablation experiments verify the effectiveness of our RS-TNet in point cloud classification and part segmentation tasks. [ABSTRACT FROM AUTHOR]