Slimming-pytorch

This project implementation of the channel pruning through BN networks slimming in Pytorch.

Code Structure

|---dataset
    |---dataset.py
|---models
    |---model.py
    |---new_resnet.py
    |---resnet_bn_slim.py
|---config.py
|---train.py
|---utils.py
|---real_prune.py

Dataset

Caltech-UCSD Birds(CUB-200-2011)

CUB-200-2011 is a bird classication task with 11,788 images from 200 wild bird species. The ratio of train data and test data is roughly 1 : 1. It is generally considered one of the most competitive datasets since each species has only 30 images for training.

Framworks

Requirements

requires python3.6, pytorch 0.40

pip install -r requirements.txt

Usage

python train.py

Reference

• Learning Efficient Convolutional Networks through Network Slimming

TODO

• Fine tune.