LightTrack: Finding Lightweight Neural Networks for Object Tracking via One-Shot Architecture Search

• DOI: 10.48550/arxiv.2104.14545
• Publication Date: 2021-01-01
• Authors: Bin Yan; Houwen Peng; Kan Wu; Dong Wang; Jianlong Fu; Huchuan Lu
• Keywords: FLOPS; BitTorrent tracker; Tracking (education); Chipset

Object tracking has achieved significant progress over the past few years. However, state-of-the-art trackers become increasingly heavy and expensive, which limits their deployments in resource-constrained applications. In this work, we present LightTrack, uses neural architecture search (NAS) to design more lightweight efficient object trackers. Comprehensive experiments show that our LightTrack is effective. It can find achieve superior performance compared handcrafted SOTA trackers, such as SiamRPN++ Ocean, while using much fewer model Flops parameters. Moreover, when deployed on mobile chipsets, discovered run faster. For example, Snapdragon 845 Adreno GPU, runs $12\times$ faster than $13\times$ parameters $38\times$ Flops. Such improvements might narrow gap between academic models industrial task. released at https://github.com/researchmm/LightTrack.