Code for Paper -- AP: Selective Activation for De-sparsifying Pruned Networks
(1) Dependencies Required:
-
Numpy
-
Pandas
-
Pytorch
-
Scikit-Learn
(2) How to Run the Code?
-
To obtain results of ResNet-20, please run resnet_20.py.
Please note that the CIFAR-10 dataset will be automatically downloaded once the code is running. The ImageNet-200 is not uploaded due to its large size. Lastly, all experimental results will be saved in a local csv file. Please check the csv file for detailed results.
(3) Description of Each Folder:
-
arch: store all different neural network architectures.
-
plots: save all plots
-
saves: save some temporal files/models
-
runs: save some results in each run
-
dumps: save some dumped files