[深度学习论文笔记][Video Classification] Beyond Short Snippets: Deep Networks for Video Classification

Hao_Zhang_Vision

于 2016-11-17 14:57:26 发布

阅读量3.3k

点赞数

分类专栏： CNN Papers 文章标签： CNN Computer Vision Deep Learning Papers Video Classification

本文链接：https://blog.youkuaiyun.com/Hao_Zhang_Vision/article/details/53200071

版权

本文探讨了深度网络在视频分类中的应用，重点比较了不同特征池化方法的效果。研究发现，最大池化由于产生更稀疏的更新，通常能更快地训练网络。在各种池化架构中，卷积池化表现最佳，保留空间信息的同时在时间域上进行池化操作至关重要。同时，使用LSTM结合光流处理图像帧可以达到最高的性能指标。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

Yue-Hei Ng, Joe, et al. “Beyond short snippets: Deep networks for video classification.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015. (Citations: 171).

1 Architecture

See Fig. We propose processing only one frame per second, and incorporate explicit motion information in the form of optical flow images computed over adjacent frames.
In all methods CNN share parameters across frames. We found that initializing from a model trained on raw image frames can help classify optical flow images by allowing

faster convergence than when training from scratch.