-
Notifications
You must be signed in to change notification settings - Fork 315
Open
Description
Hi, I am trying to convert ros CompressedImages directly to pytorch tensors for inference. Previously I the built in ros cv_bridge to covert into np array, then convert torch tensor on the gpu. However I believe cv_bridge only uses cpu decoding, and I would like to decode the images directly to the gpu. jetson has a dedicated nvjpeg hardware to do jpeg decoding, so ideally I would like to use nvjpeg to do the decoding. Are there any functions in jetson-utils that can accomplish this? the closest thing I found was loadimage, but It appears like it does not use nvjpeg.
Metadata
Metadata
Assignees
Labels
No labels