EdgeCompress: Coupling Multidimensional Model Compression and Dynamic Inference for EdgeAI

2026-07-08

AI Summary

Convolutional neural networks achieve remarkable performance in image classification, yet their high computational cost hinders deployment on edge devices. To address this challenge, this work proposes EdgeCompress, a novel framework that jointly integrates dynamic input cropping with three-dimensional (depth, width, and resolution) collaborative network compression. It introduces a cascaded dynamic inference mechanism that adaptively adjusts model complexity based on input difficulty. Guided by a lightweight foreground predictor for cropping and a composite compression strategy informed by accuracy–computation contribution trade-offs, EdgeCompress reduces the computational cost of ResNet-50 by 48.8% on ImageNet-1K while simultaneously improving Top-1 accuracy by 0.8%. At comparable computational budgets, it outperforms the state-of-the-art HRank method by 4.1% in Top-1 accuracy.

Abstract

Convolutional neural networks (CNNs) have demonstrated encouraging results in image classification tasks. However, the prohibitive computational cost of CNNs hinders the deployment of CNNs onto resource-constrained embedded devices. To address this issue, we propose EdgeCompress, a comprehensive compression framework to reduce the computational overhead of CNNs. In EdgeCompress, we first introduce dynamic image cropping (DIC), where we design a lightweight foreground predictor to accurately crop the most informative foreground object of input images for inference, which avoids redundant computation on background regions. Subsequently, we present compound shrinking (CS) to collaboratively compress the three dimensions (depth, width, and resolution) of CNNs according to their contribution to accuracy and model computation. DIC and CS together constitute a multidimensional CNN compression framework, which is able to comprehensively reduce the computational redundancy in both input images and neural network architectures, thereby improving the inference efficiency of CNNs. Further, we present a dynamic inference framework to efficiently process input images with different recognition difficulties, where we cascade multiple models with different complexities from our compression framework and dynamically adopt different models for different input images, which further compresses the computational redundancy and improves the inference efficiency of CNNs, facilitating the deployment of advanced CNNs onto embedded hardware. Experiments on ImageNet-1K demonstrate that EdgeCompress reduces the computation of ResNet-50 by 48.8% while improving the top-1 accuracy by 0.8%. Meanwhile, we improve the accuracy by 4.1% with similar computation compared to HRank, the state-of-the-art compression framework. The source code and models are available at https://github.com/ntuliuteam/edge-compress

Problem

• EdgeAI
• Model Compression
• Dynamic Inference
• Computational Overhead
• CNN Deployment

Innovation

• dynamic image cropping
• compound shrinking
• multidimensional compression
• dynamic inference
• EdgeAI