| 大会名称 |
| 2020年 総合大会 |
| 大会コ－ド |
| 2020G |
| 開催年 |
|
2020
|
| 発行日 |
|
2020-03-03
|
| セッション番号 |
|
D-11
|
| セッション名 |
|
画像工学
|
| 講演日 |
|
2020/3/20
|
| 講演場所(会議室等) |
|
工学部 講義棟２F 219講義室
|
| 講演番号 |
|
D-11-20
|
| タイトル |
|
Transfer Rate Estimation in Edge-Cloud Neural Network Solution for Object Detection
|
| 著者名 |
|
◎△Libo Hu, Tao Wang, Yucheng Zhou, Hiroshi Watanabe, Shohei Enomoto, Xu Shi, Akira Sakamoto, Takeharu Eda,
|
| キーワード |
|
quantization, branchynet, edge-cloud, exit-point, yolo, darknet
|
| 抄録 |
|
Edge devices operate not only to acquire images but also to recognize specific objects. However, edge only approach cannot take full advantage of the cloud’s cognitive capabilities. Edge-cloud cooperative approach has been proposed to solve this problem. Data of feature map should be transferred from edge to cloud. When the number of edges is large, the transfer rate becomes a bottleneck. When edge can recognize specific objects by itself, transfer rate can be reduced. When edge has no confidence for recognition, the feature maps from the branch exits will be quantized and sent to cloud.
|
| 本文pdf |
|
PDF download
|
