提出新型残差特征复用结构，提升图像分类精度与效率
Residual Feature-Reutilization Inception Network for Image Classification
- 设计残差特征复用的多尺度卷积模块，增强特征提取能力
- 在CIFAR10/100、Tiny ImageNet上分别达97.94%/85.91%/70.54%准确率
- 可调节输入分割比，减少参数量同时保持高性能
有效捕捉特征信息对计算机视觉至关重要。随着卷积神经网络的发展，残差连接与多尺度机制持续推动深度学习视觉任务性能提升。本文提出一种新型CNN架构，包含残差特征复用膨胀模块（ResFRI）或分段残差特征复用膨胀模块（Split-ResFRI）。该架构由四种不同结构的卷积组合构成，通过特制的信息交互通路连接，以提取多尺度特征并有效扩展模型感受野。此外，基于上述结构，Split-ResFRI可调整输入信息分割比例，从而减少参数量并保证模型性能。在CIFAR10（97.94%）、CIFAR100（85.91%）和Tiny ImageNet（70.54%）等主流数据集上的实验表明，该模型在相近模型规模且不使用额外数据的前提下，达到当前最优结果。
原文摘要 · Abstract (English)
Capturing feature information effectively is of great importance in the field of computer vision. With the development of convolutional neural networks (CNNs), concepts like residual connection and multiple scales promote continual performance gains in diverse deep learning vision tasks. In this paper, we propose a novel CNN architecture that it consists of residual feature-reutilization inceptions (ResFRI) or split-residual feature-reutilization inceptions (Split-ResFRI). And it is composed of four convolutional combinations of different structures connected by specially designed information interaction passages, which are utilized to extract multi-scale feature information and effectively increase the receptive field of the model. Moreover, according to the network structure designed above, Split-ResFRI can adjust the segmentation ratio of the input information, thereby reducing the number of parameters and guaranteeing the model performance. Specifically, in experiments based on popular vision datasets, such as CIFAR10 ($97.94$\%), CIFAR100 ($85.91$\%) and Tiny Imagenet ($70.54$\%), we obtain state-of-the-art results compared with other modern models under the premise that the model size is approximate and no additional data is used.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。