Creating Lightweight Object Detectors with Model Compression for Deployment on Edge Devices

TL;DR

This paper presents a novel model compression pipeline combining channel pruning, fixed channel deletion, and knowledge distillation to create lightweight object detectors suitable for edge device deployment, achieving competitive performance.

Contribution

Introduces an integrated compression pipeline that effectively reduces model size and computation for object detectors, outperforming existing lightweight models.

Findings

Resnet50-v1d pruned and fine-tuned on ImageNet

SSD-300 with 16.3MB size and 71.2 mAP

Better performance than SSD-300-MobileNet

Abstract

To achieve lightweight object detectors for deployment on the edge devices, an effective model compression pipeline is proposed in this paper. The compression pipeline consists of automatic channel pruning for the backbone, fixed channel deletion for the branch layers and knowledge distillation for the guidance learning. As results, the Resnet50-v1d is auto-pruned and fine-tuned on ImageNet to attain a compact base model as the backbone of object detector. Then, lightweight object detectors are implemented with proposed compression pipeline. For instance, the SSD-300 with model size=16.3MB, FLOPS=2.31G, and mAP=71.2 is created, revealing a better result than SSD-300-MobileNet.