Skip to content

Latest commit

 

History

History
23 lines (20 loc) · 1.28 KB

README.md

File metadata and controls

23 lines (20 loc) · 1.28 KB
Raw

Request

  • python >= 3.6.0
  • pytorch >= 0.4.0
  • numpy >= 1.14
  • cuda >= 9.0
  • torchvision >= 0.2.1

Pretrained Model

We put our trained model in the path ./data/ and you can use them directly. To repeat our work with the code, please follow these steps:

  1. Train our AlexNet with Cifar-10 dataset. Please put the dataset directory "cifar-10-batches-py" into the "data" directory. Then run the command: python cifar_alex.py
  2. Train the reconstuct net. If you want to train the reconstuct_net, please make sure the features are extracted from the fourth hidden layer of the AlexNet. And if you want to train the reconstruct_net2, please make sure the features are extracted from the fifth hidden layer. We are sorry for the inconvenient because you have to adjust this in the code of "cifar_alex.py". Then run the command: python reconstruct.py or python reconstruct_v2.py These two nets will serve as a control group in out project.

Compress Model

To apply the ADMM algorithm to the reconstruct net, please run the command: python reconstruct_pruned_new.py or python reconstructv2_pruned_new.py or python reconstructv2_pruned.py (this one partly compress the model) In the end, you can run python showResult.py to visualize the inverted images.