Residual Feature-Reutilization Inception Network for Image Classification

TL;DR

This paper introduces a novel CNN architecture called Residual Feature-Reutilization Inception Network (ResFRI) that enhances multi-scale feature extraction and increases receptive fields, achieving state-of-the-art image classification results.

Contribution

The paper proposes a new CNN design with residual feature-reutilization inception modules and a split version to optimize parameter efficiency and performance.

Findings

Achieved 97.94% on CIFAR10

Achieved 85.91% on CIFAR100

Achieved 70.54% on Tiny ImageNet

Abstract

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.