Edge and Cloud Symbiosis for Video Stream Analytics.
If you find our paper or code useful to your research, please do cite our article.
@inproceedings{nigade2020clownfish,
title={{Clownfish}: Edge and Cloud Symbiosis for Video Stream Analytics,
author={Vinod Nigade and Lin Wang and Henri Bal},
booktitle={{ACM/IEEE} Symposium on Edge Computing ({SEC})},
year={2020}
}Clone this repository and checkout submodules.
$ git submodule update --init --recursive
This repository contains three submodules.
This submodule contains code to train (fine-tune) models such as 3D-ResNets, SimiNet, EarlyDiscard (3D-MobileNet), etc. on the PKU-MMD action recognition dataset. It also contains scripts to dump prediction scores (results) in JSON files for videos using a sliding window scheme.
This submodule contains scripts to analyze the accuracy of fusion for different filter intervals and lag values. It uses the prediction scores (results) of edge and cloud models which are dumped using the above-mentioned submodule (clownfish-3D-models).
This submodule implements the Clownfish prototype. It can be used to run Clownfish components on an edge and cloud node, for action recognition task on the PKU-MMD dataset.
Please refer to the individual README files for more details.