Closed rocketmax closed 4 years ago
Hi @rocketmax,
I am sorry to hear about the issue. As you said already, you are using a GTX 960M (2 GB) graphics card. It is evident from the following log message that you are running out of memory-
Check failed: error == cudaSuccess (2 vs. 0) out of memory
This package, i.e. ros_openpose doesn't use GPU directly. However, it depends on OpenPose which uses GPU heavily.
Since the error is coming from OpenPose software, I suggest you run the openpose software on example data as shown below-
./build/examples/openpose/openpose.bin --video examples/media/video.avi
Now, please try to reduce the --net_resolution
and give it a try with --model_pose COCO.
Also, please make sure that cuDNN
is installed correctly. You need to compile OpenPose with it.
I also suggest you read and try this documentation.
Let me know if this helps!
Hey @ravijo that helped a lot! The COCO model does use less memory and works well for me. Now I'm having an issue similar to Issue #4 , but I am going to keep working on it.
Thanks again!
Hi @rocketmax,
Great! Glad that it worked. Feel free to post your error if you can't fix it.
Hello,
I know that I'm pushing my luck trying to run this on a laptop with a GTX 960M (2 GB), but I'm fairly limited in my hardware right now. The Openpose demo will run if I lower the net_resolution parameter down below -1x224. I've tried to run this with that parameter added to the openpose_args line in the launch file:
<arg name="openpose_args" value=" --model_folder /home/ubuntafoo/openpose/models/ --net_resolution -1x160 --model_pose BODY_25" />
I've even tried to lower it down to -1x16 and that didn't help, so I'm wondering if there is something within the system that locks the network resolution or if there is any way to lower the graphics memory needed that you know of.
Thanks for any help you can provide!
Here's my full error log