Image retrieval targets to find images from a database that are visually similar to the query image. Two-stage methods following retrieve-and-rerank paradigm have achieved excellent performance, but their separate local and global modules are inefficient to real-world applications. To better trade-off retrieval efficiency and accuracy, some approaches fuse global and local feature into a joint representation to perform single-stage image retrieval. However, they are still challenging due to various situations to tackle, e.g., background, occlusion and viewpoint. In this work, we design a Coarse-to-Fine framework to learn Compact Discriminative representation (CFCD) for end-to-end single-stage image retrieval-requiring only image-level labels. Specifically, we first design a novel adaptive softmax-based loss which dynamically tunes its scale and margin within each mini-batch and increases them progressively to strengthen supervision during training and intraclass compactness. Furthermore, we propose a mechanism which attentively selects prominent local descriptors and infuse fine-grained semantic relations into the global representation by a hard negative sampling strategy to optimize inter-class distinctiveness at a global scale. Extensive experimental results have demonstrated the effectiveness of our method, which achieves state-of-the-art single-stage image retrieval performance on benchmarks such as Revisited Oxford and Revisited Paris. Code is available at https://github.com/bassyess/CFCD.

Graph convolutional networks (GCNs) have proven their effectiveness and success in skeleton-based action recognition. Regular GCNs separate spatial and temporal operations. And for spatial graph convolution, the self-attention mechanism is commonly used to enable the generation of adaptive data-driven graphs, which limits the receptive field of each vertex. In order to address such a concern, we propose the graph involution network (GIN). A novel operation called graph involution is applied to offer each joint a global reception field, while also largely improving temporal-wise and channel-wise flexibility during the process of feature aggregation. We also find that predefined graphs that came from human natural skeleton structures restrict the remote information exchanges. Therefore we also propose a novel graph topology called spiral graph topology, connecting body joints spirally, which brings valuable additional information hints and improves feature aggregation efficiency. Experiments conducted on different benchmarks on two large-scale datasets show the effectiveness of our proposed network.

3D modeling is a popular topic in the field of computer vision. Most of the existing methods based on the interest point matches or the line matches. The point-based methods are more mature, but usually less intuitive. Lines can provide more structural information and thus more intuitive, however, lines have more complex geometric properties, which makes the line-based methods usually time consuming. In this paper, we proposed a simple and effective approach of 3D modeling, which takes the intersections of the real lines and the virtual lines passing through interest points to construct the 3D models, and use a simple invariant to filter out wrong matches. Thus we can use the mature technology of point-based methods while the 3D point cloud maintain the structural information of lines. Experiments show our method can build meaningful 3D point cloud efficiently.