Importance Estimation for Neural Network Pruning

This repository contains unofficial TensorFlow 2.x based implementation of Importance Estimation for Neural Network Pruning (CVPR, 2019)

This is re-write of PyToch 1.0 based implementation available here. Unlike the official code, this repository only implements Taylor Gate, which corresponds to Gate after BN in the paper (Best method in the paper).

Environment

• Tensorflow 2.x

Getting started

1. Download ImageNet

You can download ImageNet automatically if you use huggingface-cli login and run the code.

2. Train the Model

python main.py

Result