Error in function 'matting_laplacian' about deep-photo-styletransfer HOT 2 CLOSED

Stupid question, but have you generated the Laplacian matrices with Matlab / Octave – are they saved in /gen_laplacian/Input_Laplacian_3x3_1e-7_CSR[Image-Index].mat?

Was just guessing because line of 291 in deepmatting_seg.lua:
local grad = cuda_utils.matting_laplacian(output01, h, w, CSR_rowIdx, CSR_colIdx, CSR_val, N)

But that is probably not the issue, since line 117 loads the .mat files:
local CSR_fn = 'gen_laplacian/Input_Laplacian_'..tostring(params.patch)..'x'..tostring(params.patch)..'_1e-7_CSR' .. tostring(index) .. '.mat'
If the .mat files were missing, you would already get stuck at this line.

Could you check your makefile if the paths / flags are correct? When i first compiled the file i didn't get any errors, but when i tried to run the code it crashed. The fix was adding -lluajit to LDFLAGS_NVCC, but this is only the case for OSX.

Since the docker version is failing (which has a working torch install + cuda drivers) there has to be something wrong with the cuda_utils.cu, or maybe your card/drivers/CUDA install files?

Which card do you use? The makefile has -arch sm_35 so i'm not sure if older compute architectures are supported.

from deep-photo-styletransfer.

Wow! Brilliant! I removed -arch sm_35 and it've helped!
My card GeForce GTX 660 and that is the problem)
Thank you very much!

from deep-photo-styletransfer.