ID 原文 译文
56898 有效的模型先验(prior)能够降低模型训练对样本的需求. Modelprior, e. g. , the feature embedding, initialization, and configuration, is the key to the few-shot learning.
56899 本文基于元学习(meta learning)框架,从相关的、类别不同的数据中学习模型先验,并将这种先验应用于新类别的少样本任务. Thisstudy metalearns such prior from seen classes and apply the learned prior over few-shot task on unseen classes.
56900 与此同时,本文提出"模型组合先验"(MCP, model composition prior)方法,通过目标函数的最优条件对模型结构进行分解,并分别估计模型的各个组成部分,得到有效的分类器.这种分解方式具有较高的可解释性,能够指导在不同小样本任务中"共享"与"独立"的成分,从而指导元学习的具体实现.在人造数据中,本文方法能够恢复出小样本任务之间的关联性; Meanwhile, based on the first order optimal condition of the objective, the model composition prior (MCP) isstressed to decompose the model prior and estimate each component. The composition strategy improves theexplainability, while guiding the shared and specific parts among those few-shot tasks.
56901 在图像数据上, MCP方法能取得比当前主流方法更优异的效果. We verify the ability ofour approach to recover task relationship over the synthetic dataset, and our MCP method achieves better resultson two benchmark datasets (MiniImageNet and CUB).
56902 针对三维点云形状修复补全中难以保持形状精细结构信息的问题,借助于生成对抗网络框架,本文提出了一种自动修复补全三维点云形状的神经网络结构. Due to the difficulty in maintaining the fine structures of 3D point cloud in shape completion, thisstudy, with the help of the generative adversarial network framework, proposes a novel neural network for au?tomatically repairing and completing the 3D shape of point clouds.
56903 该网络由生成器和判别器构成. This network consists of a generator and adiscriminator.
56904 神经网络的生成器采用编码器–解码器结构,以缺失的三维点云形状作为输入,首先通过输入变换和特征变换对齐输入点云数据的采样点位置与特征信息; The generator of the proposed neural network adopts an encoder-decoder structure and takes themissing 3D point cloud shape data as the input. Firstly, it aligns the sampling point position and feature infor?mation of the input point cloud data by the input transform and feature transform.
56905 然后借助权共享多层感知器对各采样点提取局部形状特征并利用最大池化层与多层感知器编码提取出采样点的特征码字; Then the weighted sharedmulti-layer perceptron extracts the local shape features for each sampling point and also extracts its feature code?words using the maximum pool layer and multi-layer perceptron coding.
56906 其次将采样点特征码字加上网格坐标数据,解码器使用2个连续的三层感知器折叠操作将网格数据转变成点云形状的缺失补全数据; Secondly, it adds the feature codewordsof sampling points with the grid coordinate data, and the decoder converts the grid data into the missing data ofthe underlying point cloud using two successive three-layer perceptron folding operations.
56907 最后将缺失补全数据与点云输入数据合并,得到完整的三维点云形状. Finally, it merges themissing completion data and the input data to get the complete 3D point cloud shape.