STResNet & STYOLO : A New Family of Compact Classification and Object Detection Models for MCUs
Sudhakar Sah, Ravish Kumar
TL;DR
This paper tackles the challenge of running accurate neural networks on memory- and compute-constrained edge devices by introducing STResNet for image classification and STYOLO for object detection, both built with CompressNAS, a framework that combines layer-wise low-rank decomposition with ILP-guided channel optimization while preserving a ResNet-like backbone. By applying Tucker decomposition per layer and optimizing ranks under hardware constraints, the authors achieve three- to twelve-fold model compression with minimal accuracy loss, and they demonstrate hardware-friendly performance on STM32N6 MCUs/NPUs. STResNet variants reach about 70–72% Top-1 accuracy on ImageNet with under 4 million parameters, and STYOLO variants achieve competitive MS COCO mAP scores (e.g., ~30–34% for Micro/Milli) while maintaining small footprints; further gains are realized through a RAM-efficient projection layer and layer-wise LR tuning. The results, including Ultralytics experiments with pretrained backbones, show superior edge-device performance relative to lightweight baselines (e.g., MobileNets, YOLOv5n) and competitive proximity to larger detectors, indicating strong practical impact for on-device vision applications.
Abstract
Recent advancements in lightweight neural networks have significantly improved the efficiency of deploying deep learning models on edge hardware. However, most existing architectures still trade accuracy for latency, which limits their applicability on microcontroller and neural processing unit based devices. In this work, we introduce two new model families, STResNet for image classification and STYOLO for object detection, jointly optimized for accuracy, efficiency, and memory footprint on resource constrained platforms. The proposed STResNet series, ranging from Nano to Tiny variants, achieves competitive ImageNet 1K accuracy within a four million parameter budget. Specifically, STResNetMilli attains 70.0 percent Top 1 accuracy with only three million parameters, outperforming MobileNetV1 and ShuffleNetV2 at comparable computational complexity. For object detection, STYOLOMicro and STYOLOMilli achieve 30.5 percent and 33.6 percent mean average precision, respectively, on the MS COCO dataset, surpassing YOLOv5n and YOLOX Nano in both accuracy and efficiency. Furthermore, when STResNetMilli is used as a backbone with the Ultralytics training environment.