Publication

This work, RANP, is accepted as an oral paper by 3DV 2020 and awarded "Best Student Paper". If you find our paper or code useful, please cite it below.

@article{xu2020ranp,
title={RANP: Resource Aware Neuron Pruning at Initialization for 3D CNNs},
author={Zhiwei Xu, Thalaiyasingam Ajanthan, Vibhav Vineet, and Richard Hartley},
journal={Internatinoal Conference on 3D Vision},
year={2020}
}

Demo

This repository will include demos of neuron pruning on 3D-UNets for 3D semantic segmentation as well as MobileNetV2 and I3D for video classification.

Environment

Dependency

conda create -n RANP python=3.6.12
source activate RANP
conda install pytorch=1.1.0 torchvision cudatoolkit=9.0 -c pytorch
conda install -c conda-forge tensorboardx
conda install -c anaconda scipy==1.3.2
conda install -c conda-forge nibabel==3.2.1
conda install -c conda-forge nilearn==0.7.0
conda install -c anaconda pytables==3.4.4
conda install -c anaconda opencv==3.4.2

A possible error when installing cv2 for UCF101 experiments is
```
ImportError: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.22' not found
```
To check if this version is in the conda RANP environment, run
```
cd [path]/anaconda3/envs/RANP/lib
strings libstdc++.so.6 | grep GLIBCXX
```
If yes, add the following to ~/.bashrc, then source .bashrc; otherwise, download it, say libstdc++.so.6.0.22, first, and replace libstdc++.so.6 with a new soft-link to this downloaded dynamic library.
```
export LD_PRELOAD=[path]/anaconda3/envs/RANP/lib/libstdc++.so.6:$LD_PRELOAD
```
Datasets and proprocessing
```
cd datasets
./download_datasets.sh
```

How to use

Create subfolders below.
```
mkdir data data/shapenet data/brats data/ucf101
```
Then, download precalculated gradients from OneDrive (users can also skip this step, then running the following bash scripts will take time to generate these files automatically). Put them in the subfolders individually.
Run: configurations in the bash files are default, change the argparse parameters carefully when necessary.
- For ShapeNet experiments,
```
./run_shapenet.sh
```
- For BraTS experiments,
```
./run_brats.sh
```
- For for UCF101 experiments (set MobileNetV2 or I3D in this file),
```
./run_ucf101.sh
```

Notes

We will keep updating this repository. If you have any questions, please contact zhiwei.xu@anu.edu.au.

zwxu064/RANP

Publication

Demo

Environment

How to use

Notes