AdaptiveNet_artifact

This repo is the PyTorch implementation of the paper AdaptiveNet: Post-deployment Neural Architecture Adaptation for Diverse Edge Environments (MobiCom 2023).

Environment

The required packages of the environment we used to conduct experiments are listed in requirements.txt.

Datatset organization

You can download from datasets for test, and organize it as following:

• repo
  • datasets
    • imagenet
    • CamVid
    • coco
  • oncloud
    • mytimm
    • scripts
    • ...
  • ondevice

Experiments

for finding the optimal subnets:

cd oncloud
bash scripts/finder.sh

for training from scratch:

cd oncloud
bash scripts/train_cls.sh

Acknowledgements

We would like to thank the code from timm, EfficientDetV2, and segmentation-models.pytorch.