arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1608.06993cs.CVcs.LG
收藏

密集连接的卷积网络

Densely Connected Convolutional Networks

Cornell University(康奈尔大学);Tsinghua University(清华大学);Facebook AI Research(脸书人工智能研究院)

Gao Huang, Zhuang Liu, Laurens van der Maaten, Kilian Q. Weinberger

摘要:

最近的工作表明,如果卷积网络在靠近输入的层和靠近输出的层之间包含更短的连接,那么它们可以更深入,更准确,更有效地进行训练。在本文中,我们接受了这一观察,并介绍了密集卷积网络(DenseNet),它以前馈方式将每一层连接到其他每一层。传统的卷积网络有L层,每层和后续层之间有一个连接,而我们的网络有L(L+1)/2个直接连接。对于每个层,所有先前层的特征图被用作输入,并且其自身的特征图被用作所有后续层的输入。DenseNets有几个令人信服的优势:它们缓解了消失梯度问题,加强了特征传播,鼓励特征重用,并大大减少了参数的数量。我们在四个高度竞争的对象识别基准任务(CIFAR-10,CIFAR-100,SVHN和ImageNet)上评估了我们提出的架构。DenseNets在大多数最先进的技术上获得了显着的改进,同时需要更少的计算来实现高性能。代码和预训练模型可在https://github.com/liuzhuang13/DenseNet上获得。

英文摘要:

Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections - one between each layer and its subsequent layer - our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less computation to achieve high performance. Code and pre-trained models are available at https://github.com/liuzhuang13/DenseNet .

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑