train for fcn by caffe-windows-cudnn
northchert opened this issue · comments
The caffe-windows-cudnn can test the fcn8 well. But the big problem is how to train for the fcn8 model. The layer of input is datalayer, then it could not build lmdb file in the windows for many times. So, is there a codes for inputting images directly and successful training?
You can follow the comments from BVLC/caffe#1698 and convert the data and label lmdb.
@northchert , how did you test the fcn8? in the downloaded zip file, there is only one folder (example/fcn) related to fcn. Or did you follow the instruction from the model zoo?