Abstract Recent works have shown that convolutional neural networks (CNNs) are parameter redundant, which limits the application of CNNs in Mobile devices with limited memory and computational resources. In this paper, two novel and efficient lightweight CNNs architectures are proposed, which are called DenseDsc and Dense2Net. Two proposed CNNs are densely connected and the dense connectivity facilitates feature re-use in the networks. Dense2Net adopts efficient group convolution and DenseDsc adopts more efficient depthwise separable convolution. The novel dense blocks of DenseDsc and Dense2Net improve the parameter efficiency. The proposed DenseDsc and Dense2Net are evaluated on highly competitive classification benchmark datasets (CIFAR and ImageNet). The experimental results show that DenseDsc and Dense2Net have higher accuracy than DenseNet with similar parameters or FLOPs. Compared with other efficient CNNs with less than 0.5 M parameters for CIFAR, Dense2Net and DenseDsc achieve state-of-the-art results on CIFAR-10 and CIFAR-100, respectively. DenseDsc and Dense2Net are very competitive in efficient CNNs with less than 1.0 M parameters on CIFAR. Furthermore, Dense2Net achieves state-of-the-art results on ImageNet in manual CNNs with less than 10 M parameters.

中文翻译：

---

高效密集连接的卷积神经网络

摘要 最近的工作表明，卷积神经网络 (CNN) 是参数冗余的，这限制了 CNN 在内存和计算资源有限的移动设备中的应用。在本文中，提出了两种新颖高效的轻量级 CNN 架构，分别称为 DenseDsc 和 Dense2Net。两个提议的 CNN 是密集连接的，密集连接有利于网络中的特征重用。Dense2Net 采用高效的组卷积，DenseDsc 采用更高效的深度可分离卷积。DenseDsc 和 Dense2Net 的新型密集块提高了参数效率。提议的 DenseDsc 和 Dense2Net 在竞争激烈的分类基准数据集（CIFAR 和 ImageNet）上进行评估。实验结果表明，DenseDsc 和 Dense2Net 比具有相似参数或 FLOPs 的 DenseNet 具有更高的准确率。与 CIFAR 参数小于 0.5 M 的其他高效 CNN 相比，Dense2Net 和 DenseDsc 分别在 CIFAR-10 和 CIFAR-100 上取得了最先进的结果。DenseDsc 和 Dense2Net 在 CIFAR 上参数少于 1.0 M 的高效 CNN 中非常有竞争力。此外，Dense2Net 在手动 CNN 中的 ImageNet 上取得了最先进的结果，参数少于 10 M。