基于迁移学习和注意力机制的视频分类
首发时间:2019-12-23
摘要:受到图像分类和机器翻译的研究成果的启发,本文将其成功的体系结构设计(例如卷积神经网络和注意力机制)引入视频分类。本文尝试了不同的模型架构,包括原始的2d设计及拓展的3D结构。实验结果表明,通过使用迁移学习和注意力机制,原始的模型能够提升视频分类的准确性,其在ucf-101数据集上达到了95.5%,在hmdb-51数据集上达到了67.7%。
For information in English, please click here
Video classification based on transfer learning and attention mechanism
Abstract:Attractedby the promising result in image classification and machine translation,we introduce their successful architecture designs, such as convolutional neural networks and attention mechanism into video classification. We try different model architectures,including the original 2d design and their 3D counterpart.Experimental results demonstrate that by using transfer learning and attention techniques,our model improve accuracy in video classification problems,reaching 95.5% on ucf-101 and 67.7% on hmdb-51.
Keywords: deep learning attention mechanism transfer learning
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于迁移学习和注意力机制的视频分类
评论
全部评论0/1000