Few-shot 3D Point Cloud Object Detection Using Intention-Attention Guidance
Current point-cloud object detection methods often rely on supervised datasets,which pose challenges related to work force and resources.To address this,an innovative meta-learning framework is proposed to reduce the dependency on large labeled datasets.This research explores applying few-shot learning techniques to 3D point-cloud object detection,enabling the classification of unlabeled samples using only a few labeled examples from new classes,achieving strong performance under limited data conditions.A prototypical VoteNet is introduced to learn geometric prototypes of categories and support set prototypes.An intention-attention mechanism is also employed to capture point-cloud contextual information for more precise information fusion.Mean pooling is applied to mitigate overreliance on max-pooling,preventing the loss of critical information during prototype generation.Compared with baseline models on benchmark datasets,the proposed method consistently demonstrated significant improvements,underscoring its potential for further research and practical applications in point-cloud object detection.
point cloudtarget detectionfew-shotVoteNetintention-attention
万方数据
维普
