implement new python loss layer error

[elysion122]

Issue summary

I try implement a python loss layer in video-caffe-refactor, but it seems that the blob in python layer is still 4-N. The python Layer is like this: class MSSSIM(caffe.Layer): "A loss layer that calculates (1-MSSSIM) loss. Assuming bottom[0] is output data and bottom[1] is label, meaning no back-propagation to bottom[1]."

def setup(self, bottom, top):
    params = eval(self.param_str)
    self.C1 = params.get('C1', 0.01) ** 2
    self.C2 = params.get('C2', 0.03) ** 2
    self.sigma = params.get('sigma', (0.5, 1., 2., 4., 8.))

    # check input pair
    if len(bottom) != 2:
        raise Exception("Need two inputs to compute distance.")

    if (bottom[0].width%2) != 1 or (bottom[1].width%2) != 1 :
        raise Exception("Odd patch size preferred")

.....

the error is like this:

Creating layer loss I1208 15:46:54.000308 22662 net.cpp:100] Creating Layer loss I1208 15:46:54.000342 22662 net.cpp:434] loss <- sum I1208 15:46:54.000352 22662 net.cpp:434] loss <- label-1 I1208 15:46:54.000362 22662 net.cpp:408] loss -> loss {'alpha':0.1} F1208 15:46:54.000448 22662 blob.hpp:140] Check failed: num_axes() <= 4 (5 vs. 4) Cannot use legacy accessors on Blobs with > 4 axes. Check failure stack trace: @ 0x7f0030c83daa (unknown) @ 0x7f0030c83ce4 (unknown) @ 0x7f0030c836e6 (unknown) @ 0x7f0030c86687 (unknown) @ 0x7f00312bec1c caffe::Blob<>::LegacyShape() @ 0x7effee812cc6 boost::python::objects::caller_py_function_impl<>::operator()() @ 0x7f002fb5b64a (unknown) @ 0x7f002fb5b9b8 (unknown) @ 0x7f002fb65c93 (unknown) @ 0x7f002fb5a2c3 (unknown) @ 0x7f002f783dc3 PyObject_Call @ 0x7f002f78708c PyObject_CallFunction @ 0x7f002f7cf0cb _PyObject_GenericGetAttrWithDict @ 0x7f002f832765 PyEval_EvalFrameEx @ 0x7f002f8381ce PyEval_EvalCodeEx @ 0x7f002f7b37e1 function_call @ 0x7f002f783dc3 PyObject_Call @ 0x7f002f79654f instancemethod_call @ 0x7f002f783dc3 PyObject_Call @ 0x7f002f82dd63 PyEval_CallObjectWithKeywords @ 0x7f002f853aca PyEval_CallFunction @ 0x7effee81c3d4 caffe::PythonLayer<>::LayerSetUp() @ 0x7f0031407a66 caffe::Net<>::Init() @ 0x7f0031408905 caffe::Net<>::Net() @ 0x7f00312aecfa caffe::Solver<>::InitTrainNet() @ 0x7f00312afdfc caffe::Solver<>::Init() @ 0x7f00312b012a caffe::Solver<>::Solver() @ 0x7f003145cf43 caffe::Creator_SGDSolver<>() @ 0x40f56d caffe::SolverRegistry<>::CreateSolver() @ 0x408664 train() @ 0x405f3c main @ 0x7f002f180ec5 (unknown)

How can I use python loss layer with 5-N blob？ Thank you.