Towards Real-Time Segmentation on the Edge

• DOI: 10.1609/aaai.v37i2.25232
• Publication Date: 2023-06-27T16:12:48Z
• Authors: Yanyu Li; Changdi Yang; Pu Zhao; Geng Yuan; Wei Niu; Jiexiong Guan; Hao Tang; Minghai Qin; Qing Jin; Bin Ren; Xue Lin; Yanzhi Wang

The research in real-time segmentation mainly focuses on desktop GPUs. However, autonomous driving and many other applications rely the edge, current arts are far from goal. In addition, recent advances vision transformers also inspire us to re-design network architecture for dense prediction task. this work, we propose combine self attention block with lightweight convolutions form new building blocks, employ latency constraints search an efficient sub-network. We train MLP model based generated configurations their measured mobile devices, so that can predict of subnets during phase. To best our knowledge, first achieve over 74% mIoU Cityscapes semi-real-time inference (over 15 FPS) GPU off-the-shelf phone.