Inference time issue

[henry890112]

When I run python3 contact_graspnet/inference_ros.py --local_regions --filter_grasps and use my own RGB and depth data from the RealSense D435 camera (resolution: 640*480) on my computer with an Nvidia GeForce RTX 3060 GPU (12GB), the output is satisfactory. However, I am experiencing an inference time of approximately 45 seconds, whereas I noticed in your paper that the reported inference time is 0.28 seconds. Could you please help me identify the issue I might be encountering?
Thank you!!!

[MartinSmeyer]

Hey,
Have you checked other issues? There seems to be an issue with 30xx GPUs and the used TF/Cuda version. You can try the fixes proposed here:

#9 (comment)