当前位置:网站首页>Transfer learning analysis (CS CV) of convolutional neural network in art images

Transfer learning analysis (CS CV) of convolutional neural network in art images

2020-12-07 19:16:53 Ling Qian

Transfer learning from huge natural image datasets , Fine tuning of deep neural networks and the use of corresponding pre training networks have become the core of the de facto art analysis applications . However , Little is known about the impact of transfer learning . In this paper , We first used the technology of visualizing the internal representation of the network , To provide clues , In order to understand what art image the Internet has learned . then , Through the measurement of feature space and parameter space and the measurement calculated on the maximum active image set , Quantitative analysis of the changes brought about by the learning process . These analyses are based on several changes in the process of transfer learning . especially , We have observed that the network can use some pre trained filters specifically for new image modes , And higher layers tend to concentrate classes . Last , We show that double fine tuning of a medium-sized art dataset can improve classification on a small dataset , Even if the task changes .

Original title :An analysis of the transfer learning of convolutional neural networks for artistic images

original text :Transfer learning from huge natural image datasets, fine-tuning of deep neural networks and the use of the corresponding pre-trained networks have become de facto the core of art analysis applications. Nevertheless, the effects of transfer learning are still poorly understood. In this paper, we first use techniques for visualizing the network internal representations in order to provide clues to the understanding of what the network has learned on artistic images. Then, we provide a quantitative analysis of the changes introduced by the learning process thanks to metrics in both the feature and parameter spaces, as well as metrics computed on the set of maximal activation images. These analyses are performed on several variations of the transfer learning procedure. In particular, we observed that the network could specialize some pre-trained filters to the new image modality and also that higher layers tend to concentrate classes. Finally, we have shown that a double fine-tuning involving a medium-size artistic dataset can improve the classification on smaller datasets, even when the task changes.

Original author :Nicolas Gonthier, Yann Gousseau, Saïd Ladjal

Original address :https://arxiv.org/abs/2011.02727

Original statement , This article is authorized by the author + Community publication , Unauthorized , Shall not be reproduced .

If there is any infringement , Please contact the yunjia_community@tencent.com Delete .

版权声明
本文为[Ling Qian]所创,转载请带上原文链接,感谢
https://chowdera.com/2020/11/202011121825454394.html