Cross-Task Benchmarking of CNN Architectures

TL;DR

This study compares various dynamic CNN architectures across multiple tasks, demonstrating that attention mechanisms and adaptive convolutions improve accuracy and efficiency over traditional CNNs, with ODConv excelling on complex images.

Contribution

It introduces a comprehensive comparison of dynamic CNN variants, highlighting their advantages in accuracy and adaptability across diverse data modalities and tasks.

Findings

Attention mechanisms improve CNN accuracy.

Dynamic convolutions enhance efficiency.

ODConv excels on complex images.

Abstract

This project provides a comparative study of dynamic convolutional neural networks (CNNs) for various tasks, including image classification, segmentation, and time series analysis. Based on the ResNet-18 architecture, we compare five variants of CNNs: the vanilla CNN, the hard attention-based CNN, the soft attention-based CNN with local (pixel-wise) and global (image-wise) feature attention, and the omni-directional CNN (ODConv). Experiments on Tiny ImageNet, Pascal VOC, and the UCR Time Series Classification Archive illustrate that attention mechanisms and dynamic convolution methods consistently exceed conventional CNNs in accuracy, efficiency, and computational performance. ODConv was especially effective on morphologically complex images by being able to dynamically adjust to varying spatial patterns. Dynamic CNNs enhanced feature representation and cross-task generalization through adaptive kernel modulation. This project provides perspectives on advanced CNN design architecture for multiplexed data modalities and indicates promising directions in neural network engineering.