Dentists judge the quality of root canal therapy for each patient very time-consuming, and inefficient, lack of quantitative evaluation criteria, easy to cause judgment errors. At the same time, the traditional method...
详细信息
This paper proposes a novel short-term building load forecasting approach under the framework of patch learning, a novel data-driven model that aggregates a global model and several patch models to further reduce fore...
详细信息
Towards open-ended Video Anomaly Detection (VAD), existing methods often exhibit biased detection when faced with challenging or unseen events and lack interpretability. To address these drawbacks, we propose Holmes-V...
详细信息
Battery energy storage systems are widely used in microgrids integrated with volatile energy resources for their ability in peak load shifting. Security constrained economic dispatch over the system’s lifecycle is a ...
详细信息
Transferring vision-language knowledge from pretrained multimodal foundation models to various downstream tasks is a promising direction. However, most current few-shot action recognition methods are still limited to ...
详细信息
Point-supervised Temporal Action Localization (PSTAL) is an emerging research direction for label-efficient learning. However, current methods mainly focus on optimizing the network either at the snippet-level or the ...
详细信息
Partially annotated images are easy to obtain in multi-label classification. However, unknown labels in partially annotated images exacerbate the positive-negative imbalance inherent in multi-label classification, whi...
Partially annotated images are easy to obtain in multi-label classification. However, unknown labels in partially annotated images exacerbate the positive-negative imbalance inherent in multi-label classification, which affects supervised learning of known labels. Most current methods require sufficient image annotations, and do not focus on the imbalance of the labels in the supervised training phase. In this paper, we propose saliency regularization (SR) for a novel self-training framework. In particular, we model saliency on the class-specific maps, and strengthen the saliency of object regions corresponding to the present labels. Besides, we introduce consistency regularization to mine unlabeled information to complement unknown labels with the help of SR. It is verified to alleviate the negative dominance caused by the imbalance, and achieve state-of-the-art performance on Pascal VOC 2007, MS-COCO, VG-200, and Openimages V3.
Although deep learning of 3D point clouds has made significant progress, the robustness of 3D models has not been fully investigated. Existing 3D attack methods show satisfactory performance under the white-box settin...
详细信息
Although deep learning of 3D point clouds has made significant progress, the robustness of 3D models has not been fully investigated. Existing 3D attack methods show satisfactory performance under the white-box setting but frequently suffer from a low transferability to attack black-box models. A key point to improve the transferability of attacks is to find the commonalities of multiple models. To this end, a novel attack method is proposed by only leveraging perturbations in local geometric feature areas termed FAP. Specifically, local perturbations are pertinently added by calculating the curvature to sample more points in the feature areas that 3D models potentially focus on. Moreover, to address the drawback of poor isometric robustness common to 3D models, we propose a generalized framework for diversifying input samples by adding random isometric transformations (DIT) in the attack process, which can be combined with other attack methods to improve their transferability. Evaluated on a variety of typical 3D models, the proposed attack method outperforms the current gradient-based methods in transferable black-box attacks, and the effectiveness of the proposed framework is demonstrated by extending it to other attack methods.
Due to its high degree of customization, DNA origami provides a versatile platform with which to engineer nanoscale structures and devices. Reconfigurable nanodevices driven by DNA strand displacement accomplish the t...
详细信息
暂无评论