Scalable Cloud Service For Multimedia Analysis Based on Deep Learning
Bing‐Kun Bao, Yangyang Xiang, Lusong Li, Shuen Lyu, Harsh Munshi, Honghong Zhu · 2018
Convolutional neural networks have proved their capability over wide areas of computer vision and has lead to superlative performance in the benchmarking tasks. Efforts to train and run deep neural networks is computationally expensive and requires a cloud infrastructure equiped with appropriate hardware. In this paper, we introduce the technical details and results from our cloud service for image analysis. It includes a ConvNet over region proposals, and is able to cater the requests from a wide range of object classification and detection tasks. Our approach also encompasses an auto-scaling system to handle intensive inquiries in a short amount of time and automatically sorts the queries based on their importance. The experimental results prove that our system could offer the state-of-the-art accuracy on users' data.