Pruning
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).
Install / Use
/learn @yehuitang/PruningREADME
GAN-pruning
A Pytorch implementation for our ICCV 2019 paper, Co-Evolutionary Compression for unpaired image Translation, 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, 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, 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>