Failed with UMFPACK_ERROR_out_of_memory

[Yongcheng Jing]

Thanks for the great code. When I run the algorithm with my own high-resolution images (655 * 1280), I find that when using scipy.sparse.linalg.spsolve with scikit-umfpack as solver, it requires too much memory (larger than 128GB).
After some investigations, I found the problem might be OS dependent. However, I actually followed the instructions: my OS is Ubuntu 16.04, also the same CUDA and python version.

I wonder if anyone struggles at the same issue with me, and if there is any other solver. Thanks.

[Wesley Williams]

Try uninstalling scikit-umfpack (with pip). I sometimes found the default sparse solver could handle larger images. You will find it's considerably slower though.

[Yongcheng Jing]

@wesleyw72 Thanks! But it is far too slow, especially for high-resolution images :(

[Yijun Li]

@ycjing Thanks for your interests.

You can check this work which will reduce the N x N (N is the number of pixels) adjacent matrix W to k x k for efficiency.

[Yongcheng Jing]

@Yijunmaverick Really appreciate the help! Thanks!

[785256592]

@Yijunmaverick Really appreciate the help! Thanks!

dear@ycjing I also have this problem ,can you show me how to handle it ! very thanks!
BTW

[Yongcheng Jing]

Hi @785256592 ,

Since it has been almost two years since I worked on this project, I hardly remember the problem and the solution. Really sorry for that.

[Guoqing Hao]

@Yijunmaverick Really appreciate the help! Thanks!

dear@ycjing I also have this problem ,can you show me how to handle it ! very thanks!
BTW

In my case, simply uninstalling scikit-umfpack worked.

[doantientai]

@Yijunmaverick Really appreciate the help! Thanks!

dear@ycjing I also have this problem ,can you show me how to handle it ! very thanks!
BTW

In my case, simply uninstalling scikit-umfpack worked.

Thank you! Uninstalling scikit-umfpack worked for me too.