Wu Y.-SWu S.-SHuang TLIANG-GEE CHEN2022-04-252022-04-25202102714310https://www.scopus.com/inward/record.uri?eid=2-s2.0-85109027731&doi=10.1109%2fISCAS51556.2021.9401468&partnerID=40&md5=3e2a45d0feb3299e07abc5c0c932669ahttps://scholars.lib.ntu.edu.tw/handle/123456789/607245Sending local data to cloud servers is vulnerable to user privacy, and its long update latency. Meanwhile, the state-of-the-art stereo matching method is still computation demanding, fine-tuning the whole model on-device is not a practicable solution because of the limited power budget and computation ability on edge devices. In this study, we propose a two-stage online stereo matching refinement system, using an additional light-weight network to learn the domain gap between local data and cloud training data. We define a load-gain ratio to evaluate computer efficiency. This refinement system has a much better load-gain ratio than fine-tune. (0.2 v.s. 35.7 operation overhead/accuracy gain) Nevertheless, we only disburse 0.2% of additional parameters and 0.7% additional computation as set by inference the stereo matching model. Thus, it would be a suitable choice for an online training scenario. With re-scheduling the training pipeline, we use a patch-based layer fusion technique and reduce the off-chip memory bandwidth by 97%. ? 2021 IEEEDevice personalizationLayer fusionOnline trainingPatch-based layer fusionRefinement networkStereo matchingBudget controlOnline systemsPrivacy by designArchitecture designsComputation abilityComputer efficiencyFusion techniquesOff-chip memoryState of the artStereo matching methodE-learning[SDGs]SDG3conference paper10.1109/ISCAS51556.2021.94014682-s2.0-85109027731