Experimental Comparison of Light-Weight and Deep CNN Models Across Diverse Datasets

TL;DR

This paper compares lightweight and deep CNN models across diverse datasets, demonstrating that well-regularized shallow architectures can be competitive and practical for real-world, resource-constrained applications.

Contribution

It establishes a unified benchmark for Bangladeshi vision datasets and highlights the effectiveness of lightweight CNNs in various real-world scenarios.

Findings

Shallow CNNs perform competitively with deep models across datasets.

Lightweight models are effective without large GPU resources.

Benchmarking provides a practical reference for low-resource deployments.

Abstract

Our results reveal that a well-regularized shallow architecture can serve as a highly competitive baseline across heterogeneous domains - from smart-city surveillance to agricultural variety classification - without requiring large GPUs or specialized pre-trained models. This work establishes a unified, reproducible benchmark for multiple Bangladeshi vision datasets and highlights the practical value of lightweight CNNs for real-world deployment in low-resource settings.