search

ivalab / grasp_multiObject_multiGrasp

An implementation of our RA-L work 'Real-world Multi-object, Multi-grasp Detection'
204 stars 49 forks source link

Detection took 21.873s only for one picture? #22

Open Ultraopxt opened 3 years ago

Ultraopxt commented 3 years ago

Hi, I have run the demo, but Detection took 21.873s for 300 object proposals for only one image(640*360size). Far away from 0.25s in the paper. But I don't know reason.

Ultraopxt commented 3 years ago

Who can help?

fujenchu commented 3 years ago

what is your fps if you disable gpu then?

Ultraopxt commented 3 years ago

what is your fps if you disable gpu then?

Thanks.Detection took 19.486s for 300 object proposals without gpu. How long did you spent for an image?

fujenchu commented 3 years ago

you might want to check if the gpu is utilized properly given your gpu and cpu spend similar amount of time.

Ultraopxt commented 3 years ago

you might want to check if the gpu is utilized properly given your gpu and cpu spend similar amount of time.

Thanks. I will check

Ultraopxt commented 3 years ago

you might want to check if the gpu is utilized properly given your gpu and cpu spend similar amount of time.

I use the demo_graspRGD.py to do inference.But I notice that it looks like no code refer to gpu? This demo_graspRGD.py is only for cpu?

fujenchu commented 3 years ago

If you install the gpu version of tf properly then by default it will utilize your gpu, unless you make it invisible. This script was tested with a machine with a gpu. Hope this information helps.