论文阅读 [TPAMI-2022] Label Independent Memory for Semi-Supervised Few-Shot Video Classification
论文阅读 [TPAMI-2022] Label Independent Memory for Semi-Supervised Few-Shot Video Classification
论文搜索(studyai.com)
搜索论文: Label Independent Memory for Semi-Supervised Few-Shot Video Classification
搜索论文: http://www.studyai.com/search/whole-site/?q=Label+Independent+Memory+for+Semi-Supervised+Few-Shot+Video+Classification
关键字(Keywords)
Training; Feature extraction; Task analysis; Compounds; Dynamics; Data models; Prototypes; Few-shot video classification; semi-supervised learning; memory-augmented neural networks; compound memory networks
机器学习; 机器视觉
监督学习; 半监督学习; 小样本学习; 视频分类; 多模态感知; 相似性度量与搜索
摘要(Abstract)
In this paper, we propose to leverage freely available unlabeled video data to facilitate few-shot video classification.
在本文中，我们建议利用免费可用的未标记视频数据，以方便对少量镜头的视频进行分类。.
In this semi-supervised few-shot video classification task, millions of unlabeled data are available for each episode during training.
在这个半监督的少镜头视频分类任务中，在训练期间，每一集都有数百万未标记的数据可用。.
These videos can be extremely imbalanced, while they have profound visual and motion dynamics.
这些视频可能极不平衡，但它们具有深刻的视觉和运动动态。.
To tackle the semi-supervised few-shot video classification problem, we make the following contributions.
为了解决半监督少镜头视频分类问题，我们做出了以下贡献。.
First, we propose a label independent memory (LIM) to cache label related features, which enables a similarity search over a large set of videos.
首先，我们提出了一种标签无关存储器（LIM）来缓存标签相关的特征，它可以在大量视频中进行相似性搜索。.
LIM produces a class prototype for few-shot training.
LIM制作了一个用于少数投篮训练的课堂原型。.
This prototype is an aggregated embedding for each class, which is more robust to noisy video features.
该原型是每个类的聚合嵌入，对噪声视频特征更具鲁棒性。.
Second, we integrate a multi-modality compound memory network to capture both RGB and flow information.
其次，我们集成了一个多模态复合存储网络来捕获RGB和流信息。.
We propose to store the RGB and flow representation in two separate memory networks, but they are jointly optimized via a unified loss.
我们建议将RGB和流表示存储在两个单独的内存网络中，但它们通过统一的损耗进行联合优化。.
In this way, mutual communications between the two modalities are leveraged to achieve better classification performance.
通过这种方式，利用两种模式之间的相互通信来实现更好的分类性能。.
Third, we conduct extensive experiments on the few-shot Kinetics-100, Something-Something-100 datasets, which validates the effectiveness of leveraging the accessible unlabeled data for few-shot classification…
第三，我们在少量放炮动力学-100，某物-100数据集上进行了大量实验，这验证了利用可访问的未标记数据进行少量放炮分类的有效性。。.
作者(Authors)
[‘Linchao Zhu’, ‘Yi Yang’]