# GAN-pruning

**Repository Path**: mirrors_huawei-noah/GAN-pruning

## Basic Information

- **Project Name**: GAN-pruning
- **Description**: Code for "Co-Evolutionary Compression for Unpaired Image Translation" (ICCV 2019), "SCOP: Scientific Control for Reliable Neural Network Pruning" (NeurIPS 2020) and “Manifold Regularized Dynamic Network Pruning” (CVPR 2021).
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-09-24
- **Last Updated**: 2026-03-22

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

## GAN-pruning
A Pytorch implementation for our ICCV 2019 paper, [Co-Evolutionary Compression for unpaired image Translation](https://arxiv.org/abs/1907.10804), which proposes a co-evolutionary approach for reducing memory usage and FLOPs of generators on image-to-image transfer task simultaneously while maintains their performances.

<p align="center">
<img src="GAN-Pruning/fig/framework.PNG" width="600">
</p>

### Performance
Performance on cityscapes compared with conventional pruning method:
<p align="center">
<img src="GAN-Pruning/fig/FCN.PNG" width="600">
</p>


## SCOP
A Pytorch implementation for our NeurIPS 2020 paper, [SCOP: Scientific Control for Reliable Neural Network Pruning](https://arxiv.org/abs/2010.10732), which proposes a reliable neural network pruning algorithm by setting up a scientific control. 

<p align="center">
<img src="SCOP_NeurIPS2020/fig/framework.PNG" width="700">
</p>


### Performance
Comparison of the pruned networks with different methods on ImageNet.

<p align="center">
<img src="SCOP_NeurIPS2020/fig/imagenet.PNG" width="600">
</p>

## ManiDP
A Pytorch implementation for our CVPR 2021 paper, [Manifold Regularized Dynamic Network Pruning](https://openaccess.thecvf.com/content/CVPR2021/papers/Tang_Manifold_Regularized_Dynamic_Network_Pruning_CVPR_2021_paper.pdf), which proposes a dynamic pruning paradigm to maximally excavate network redundancy corresponding to input instances.

<p align="center">
<img src="ManiDP/fig/framework.PNG" width="700">
</p>


### Performance
Comparison of the pruned networks with different methods on ImageNet.

<p align="center">
<img src="ManiDP/fig/imagenet.PNG" width="600">
</p>