Out of memory while running demo

[vikram-gupta]

I am getting out of memory error while running the demo using the following command. I am using GeForce GTX 1070.

python ./pose_demo.py image.png --out_name=prediction

INFO:estimate_pose:Done!
F1028 23:43:02.663663 10368 syncedmem.cpp:56] Check failed: error == cudaSuccess (2 vs. 0)  out of memory
*** Check failure stack trace: ***
Aborted (core dumped)

[CamilleMaurice]

@vikram-gupta Same for me (but not same graphic card). Did you manage to fix it ?

[vikram-gupta]

@CamilleMaurice not yet !

[CamilleMaurice]

I've reinstalled it from scratch but I still face the same problem. If I find out a solution I'll let you know here. Btw, I am using anaconda2

[eldar]

Hi,

This is interesting. What is the size of image.png? If it is too big, then unfortunately it won't fit into memory.

Cheers,
Eldar.

[CamilleMaurice]

The size is 274x512 pixels in the demo. I've tried to resize it to a smaller size, without success. I've already installed deeplearning frameworks using Caffe and Cuda and this is the first time I face this error. I'm not using Debian Wheezy but Ubuntu 16.04 and I am using CUDA 8.0.61

[asnt]

Same problem on a 4GB Quadro M2000M. The problem persists whatever the input image size.
Do we need a 12GB card for running on GPU?

[CamilleMaurice]

Reading the paper I've also found out the use of 12GB GPU. Next week I might be able to try with 11GB GPU card. I wont have access to more GPU for now so that will be my last test and I'll report here my results.

[CamilleMaurice]

Here are some news, I managed to install it and it worked on the : GeForce GTX 1080 Ti 11Go.

[williammm001]

It doesn't work on RTX 2080 Super. I guess it might need a 11GB GPU card like 1080Ti.

[CamilleMaurice]

It doesn't work on RTX 2080 Super. I guess it might need a 11GB GPU card like 1080Ti.

Yes. The need for NVIDIA and CUDA is widespread. You often need them to run pose estimation algorithms. It is the same for OpenPose (you might be insterested).

[williammm001]

It doesn't work on RTX 2080 Super. I guess it might need a 11GB GPU card like 1080Ti.

Yes. The need for NVIDIA and CUDA is widespread. You often need them to run pose estimation algorithms. It is the same for OpenPose (you might be insterested).

Thanks. I used deepcut just because it is part of the MuVS. I just wonder why it requires so much GPU memory...