Hi,
I think there is a bug in datahandler.py at line 344 'shift = (sizeX - sizex)/2'.
This value should be 0, otherwise you suppose the possible offset translation starts from the upper-left corner in the image always with positive values. I guess this is not consistent with the code in cudamat_kernels.cu (kGenerateTranslationsBigVarOff function, line 795).
Jordi
Hi, I think there is a bug in datahandler.py at line 344 'shift = (sizeX - sizex)/2'. This value should be 0, otherwise you suppose the possible offset translation starts from the upper-left corner in the image always with positive values. I guess this is not consistent with the code in cudamat_kernels.cu (kGenerateTranslationsBigVarOff function, line 795). Jordi