GhostNet-Keras

Simple implementation 2020 CVPR:《GhostNet：More Features from Cheap Operations》

1.Environment:

keras=2.1.5
tensorflow=1.4.0
numpy
bunch
opencv

2.Train mnist data：

set "use_mnist_data"=1 from config/ghost_config.json, then run command: 
python train.py -c config/ghost_config.json

3.Train customer data:

set "use_mnist_data"=0 , "train_list" = path/to/train.txt, "test_list" = path/to/test.txt
train.txt and test.txt have the following format:
	
line1: path/to/image1.jpg label(0,1,2,3,4,......)
line2: path/to/image2.jpg label(0,1,2,3,4,......)
line3: path/to/image3.jpg label(0,1,2,3,4,......)
line4: path/to/image4.jpg label(0,1,2,3,4,......)
....

run commmand:python train.py -c config/ghost_config.json

4.Test

run commmand:python infer.py -c config/ghost_config.json