guide_retrainModels.txt

To train a new model for substracting background:
    1. Modify the information of bg substract in MyConfig.py in VGG folder. 
    2. Run trainBGscale.py
    3. The trained parameters will be stored in MyConfig.outputPath with file name as MyConfig.paramFileName.
    
To visualize the results of trained bg substractor:
    1. Modify the information of testing images in MyConfig.py
    2. Run testBGsub.ipynb
    
To train a new model for high-order potential (GMM and regressor):
    1. Modify the information of GMM and regression model in MyConfig.py. 
    2. Run retrainGaussian.py
    3. The trained parameters will be stored in MyConfig.net_params_path
    
To visualize the results of GMM and regressor:
    1. Modify the information for testing GMM in MyConfig.py
    2. Run testGaussian.ipynb
    
If you want to use new trained parameters to calculate high-order potential for CRF, in the function run_inference in file GaussianNet.py, modify the name of pickle file to load for params_bg, params_regression, and gaussian_params.



To train a new model for unary potential:
    1. Modify the information of UnariesNet in MyConfig.py. 
    2. Run retrainUnaries.py
    3. The trained parameters will be stored in the MyConfig.unaries_params_path
    
To visualize the results of unary term:
    1. Modify the information for testing unariesNet in MyConfig.py.
    2. Run testUnaries.ipynb
    
If you want to use new trained parameters to calculate unary potential for CRF, modify the name of pickle file to load for params_to_load and params_VGG in the __init__ function of class unariesNet in the file UnariesNet.py.