Improving Edge-AI Image Classification Through the Use of Better Building Blocks

Lucas Mohimont, Lilian Hollard, Luiz Angelo Steffenel

Abstract: Traditional CNN architectures for classification, while successful, suffer from limitations due to diminishing spatial resolution and vanishing gradients. The emergence of modular ”building blocks” offered a new approach, allowing complex feature extraction through stacked layers. Despite the popularity of models like VGG, their high parameter count restricts their use in resource-constrained environments like Edge AI. This work investigates efficient building blocks as alternatives to VGG blocks, comparing the performance of diverse blocks from well-known models alongside our proposal block. Extensive experiments across various datasets demonstrate that our proposed block surpasses established blocks like Inception v1 in terms of accuracy while requiring significantly fewer resources regarding computational cost (GFLOPs) and memory footprint (number of parameters). This showcases its potential for real-world applications in Edge AI.

External IDs:dblp:conf/closer/MohimontHS24