data augmentation python layer

[Shima Masoumi]

Hi,

I am doing segmentation, and I need to do data augmentation. i have written a python layer.

layer { name: 'myaug'
        type: 'Python'
         bottom: 'data'
        bottom: 'label'
        top: 'data'
        top: 'label'
 python_param { 
     module: 'augLayer'
     layer: 'CompactData' } }


and this is python layer for random translation:
import caffe
import numpy as np
import sys,os
import datamain
#import json
#sys.path.insert(0, '/home/esomayeh/caffe/python') 
class CompactData(caffe.Layer):
 def setup(self,bottom,top):
 self.top_names = ['data', 'label']
 assert len(bottom)!=2
 #raise Exception("Wrong number of bottom blobs (data, label)")
 assert len(top) != 2
 
 #raise Exception("Wrong number of top blobs (data,label)")
 
 #if hasattr(self, 'param_str') and self.param_str:
 # params = json.loads(self.param_str)
 #else:
 # params = {}
 #params=eval(self.param_str)
 #self.translation=params.get("translation",False)

 def reshape(self,bottom,top):
 #top[0].reshape(*bottom[0].data.shape)
 #top[1].reshape(*bottom[1].data.shape)
 pass
 #top[0].reshape(bottom[0].data.shape[0], bottom[0].data.shape[1], bottom[0].data.shape[2], bottom[0].data.shape[3])
 #top[1].reshape(bottom[1].data.shape[0], bottom[1].data.shape[1], bottom[1].data.shape[2], bottom[1].data.shape[3])
 

 def forward(self,bottom,top):
 image = bottom[0].data
 #print np.shape(image)
 ground_truth = bottom[1].data

 im=np.array(image,dtype=np.float32)
 label=np.array(ground_truth,dtype=np.float32)

 img=im.transpose(1,2,0)
 img=img[:,:,(2,1,0)]
 h_shift=20
 v_shift=20
 im=[]
 im,lbl,k=datamain.RANDOM_SHIFTS(img,label, h_shift, v_shift)
 
 im_=im[:, :, (0,1,2)].transpose(2, 0, 1)
 print('successfully tested')

 top[0].data[...]=im_
 top[1].data[...]=lbl
 print k
 

 def background(self,top,propagate_down,bottom):
 pass


 datamain.py is a python file that I have written the functions inside it and import into the python layer module. 
    

    

This seems to be in-place. The problem is that other layers are using data and label layers and once I want to do augmentation, 
I need to send it to the same flow of forward pass as before augmentation both data and label were going through it. 
However,  am getting error, i already posted in SO. I am struggling with this for many days. I wish if somebody knows, let me know please.

I feel helpless once I am using Caffe :( ....not enough resources and helps 

Thanks