Semantic Segmentation Learning Notes (IV.)--enet training problems

According to enet instructions Https://github.com/TimoSaemann/ENet/tree/master/Tutorial, training encountered the following problems

1 No module named Spatial_dropout

Enet through the Python interface to define a new layer spatial_dropout, according to the instructions to start training in the terminal, "No module named Spatial_dropout", because the Caffe,python layer is located in the folder path, Not added to the system Pythonpath, one workaround is to start the training with the Python interface:

Import sys  
caffe_root = ' home/f/caffe/'  
sys.path.insert (0,caffe_root+ ' python ')  

import Caffe
Caffe.set_device (0)
Caffe.set_mode_gpu ()
solver = Caffe. Sgdsolver ('/home/xxx/data/solver.prototxt ')
solver.solve ()

2 Drop1_0_3 when new, Img_height = = Height validation failed

The data layer of the New_height new_weight can not be deleted, according to the input image set size can be.

Layer {
  name: ' Data '
  type: ' Denseimagedata ' top
  : ' Data ' top
  : ' label '
  dense_image_data_param {
    Source: "Enet/dataset/train_fine_2columns.txt"
    batch_size:4
    shuffle:true
    new_height:512
    new_width:1024
    label_divide_factor:8
  }
}