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

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

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

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

for finding the optimal subnets:

cd oncloud
bash scripts/finder.sh

for training from scratch:

cd oncloud
bash scripts/train_cls.sh

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