training-sbd-instructions

1. Download the dataset from SBD site (http://home.bharathh.info/pubs/codes/SBD/download.html)

2. Copy it to '/home/XYZ/Desktop/DilatedFCNSegmentation/Dilated_FCN-2s_VGG16/data'

3. Open the train.prototxt and val.protoxt 


4. In data layer (first layer) change the layer: "VOCSegDataLayer" to layer:  "SBDDSegDataLayer"


4. In data layer change parameteter: \'voc_dir\' to parameteter: \'sbdd_dir\'

5. Path for \'sbdd_dir\' will be \'/home/XYZ/Desktop/DilatedFCNSegmentation/Dilated_FCN-2s_VGG16/data/benchmark_RELEASE/dataset\'

6. Change the name of the snapshot in solver.prototxt 
            from  '/home/XYZ/Desktop/DilatedFCNSegmentation/Dilated_FCN-2s_VGG16/snapshot/voctraining'
            to '/home/XYZ/Desktop/DilatedFCNSegmentation/Dilated_FCN-2s_VGG16/snapshot/sbdtraining'
7. Run solvedilatedsbd.py