Smart Scissor: Coupling Spatial Redundancy Reduction and CNN Compression for Embedded Hardware

2026-07-07

AI Summary

This work addresses the trade-off between computational efficiency and accuracy in convolutional neural networks (CNNs) for embedded deployment. Directly reducing input resolution often discards critical foreground features, degrading accuracy, while inherent redundancy in CNNs limits their efficiency on resource-constrained devices. To overcome these challenges, the authors propose Smart Scissor, a novel framework that unifies instance-aware dynamic foreground cropping with joint compression across network depth, width, and input resolution into an end-to-end optimization pipeline. A lightweight foreground predictor selectively removes redundant background regions, enabling co-optimization of architecture and input scale. On ImageNet-1K, Smart Scissor reduces the computational cost of ResNet-50 by 41.5% while improving Top-1 accuracy by 0.3%, substantially outperforming prior methods—e.g., surpassing HRank by 4.1% in accuracy under equivalent computational budgets.

Abstract

Scaling down the resolution of input images can greatly reduce the computational overhead of convolutional neural networks (CNNs), which is promising for edge AI. However, as an image usually contains much spatial redundancy, e.g., background pixels, directly shrinking the whole image will lose important features of the foreground object and lead to severe accuracy degradation. In this paper, we propose a dynamic image cropping framework to reduce the spatial redundancy by accurately cropping the foreground object from images. To achieve the instance-aware fine cropping, we introduce a lightweight foreground predictor to efficiently localize and crop the foreground of an image. The finely cropped images can be correctly recognized even at a small resolution. Meanwhile, computational redundancy also exists in CNN architectures. To pursue higher execution efficiency on resource-constrained embedded devices, we also propose a compound shrinking strategy to coordinately compress the three dimensions (depth, width, resolution) of CNNs. Eventually, we seamlessly combine the proposed dynamic image cropping and compound shrinking into a unified compression framework, Smart Scissor, which is expected to significantly reduce the computational overhead of CNNs while still maintaining high accuracy. Experiments on ImageNet-1K demonstrate that our method reduces the computational cost of ResNet50 by 41.5% while improving the top-1 accuracy by 0.3%. Moreover, compared to HRank, the state-of-the-art CNN compression framework, our method achieves 4.1% higher top-1 accuracy at the same computational cost. The codes and data are available at https://github.com/ntuliuteam/smart-scissor

Problem

• spatial redundancy
• CNN compression
• computational overhead
• edge AI
• accuracy degradation

Innovation

• dynamic image cropping
• spatial redundancy reduction
• CNN compression
• compound shrinking
• embedded AI