Rethinking the Value of Network Pruning (Pytorch) (ICLR 2019)
Branch: master
Clone or download
Latest commit d53069c Dec 23, 2018
Type Name Latest commit message Commit time
Failed to load latest commit information.
cifar Fix bug in network slimming Dec 5, 2018
LICENSE Initial Oct 11, 2018 Update Dec 24, 2018

Rethinking the Value of Network Pruning

This repository contains the code for reproducing the results, and trained ImageNet models, in the following paper:

Rethinking the Value of Network Pruning

Zhuang Liu*, Mingjie Sun*, Tinghui Zhou, Gao Huang, Trevor Darrell (* equal contribution).

ICLR 2019. Also Best Paper Award at NIPS 2018 Workshop on Compact Deep Neural Networks.

This paper shows that training the small pruned model from scratch can almost always achieve comparable or higher level of accuracy than the model obtained from the typical "training, pruning and fine-tuning" procedure. Our results suggest:

  1. Training a large, over-parameterized model is not absolutely necessary to obtain an efficient final model.
  2. Learned “important” weights of the large model are not necessarily useful for the small pruned model.
  3. The pruned architecture itself, rather than a set of inherited "important" weights, is what leads to the efficiency benefit in the final model, which suggests that some pruning algorithms could be seen as performing network architecture search.


  title={Rethinking the Value of Network Pruning},
  author={Zhuang Liu and Mingjie Sun and Tinghui Zhou and Gao Huang and Trevor Darrell},


We evaluated the following six pruning methods:

  1. L1-norm based channel pruning
  2. ThiNet
  3. Regression based feature reconstruction
  4. Network Slimming
  5. Sparse Structure Selection
  6. Non-structured weight-level pruning

For CIFAR, our code is based on pytorch-classification and network-slimming. For ImageNet, we use the official Pytorch ImageNet training code. The instructions and models are in each subfolder.

Our experiment environment is Python 3.6 & PyTorch 0.3.1.


sunmj15 at
liuzhuangthu at