a154609 / biasloss_and_skipnet_evaluation

Code for the evaluation of SkipNet model on ImageNet validation set

SkipNet

Demo for SkipNet architecture presented in ICCV 2021 paper of "Bias Loss for Mobile Neural Networks".

Requirements

for installing required packages run pip install -r requirements.txt

Usage

Pretrained models are available from Google Drive. For the testing please download and place them in the same directory as the validation script (currently testing available only for skipnet-m, code will be updated).

python validate.py --data path/to/the/dataset

Introduction to SkipNet

"Bias Loss for Mobile Neural Networks"

By Lusine Abrahamyan, Valentin Ziatchin, Yiming Chen and Nikos Deligiannis.

Approach

Performance

SkipNet beats other SOTA lightweight CNNs such as MobileNetV3 and FBNet.

|