Xavier SOFE support

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

have some SOFE qustion need support:

  1. Do we have SOFE’s manual and spec
  2. Does the SOFE accelerator support dense stereo matching
  3. The dense stereo method supported by SOFE accelerator (BM/SGM/…), and their specs (resolution, frame rate,…)
  4. The full parallax of the SOFE accelerator is supported
  5. Is there any SOFE demo video
    thanks
2

Hi denvend,

That will be supported in coming future release of JetPack, which is planned at Nov/E, there isn’t any documentation to share at this time. Please stay tuned.

Thanks

3

Hi kayccc,

That will be supported in coming future release of JetPack, which is planned at Nov/E, there isn’t any documentation to share at this time.

Was this release schedule postponed?

4

Hi dandelion1124,

Please wait for our announcement, thanks.

5

@kayccc Do you have a plan to provide SOFE support? I’m waiting two years.

7

@dandelion1124 I want to know if we can use SOFE now？

8

@22160115 Unfortunately I do not have any information.

@kayccc

Please wait for our announcement, thanks.

Let me know if there’s any progress.

9

yes, anything please. Thanks

10

It’s implemented at DeepStream SDK, please refer to Gst-nvof — DeepStream 6.2 Release documentation (nvidia.com)

1 Like
13

@kayccc Thanks, I’ve seen vpi using nvenc to calculate dense optical flow, is there any difference between vpi and deepstream? Looks like they’re all producing 1/16 optical flow of the input image, right?

15

Hi,

They use the same hardware to get the optical flow result.

However, the library is different.
Deepstream is preferred if you want to combine it with some deep learning inference.
VPI is suitable for integrating with other vision based algorithms.

Thanks.

1 Like
16

got it， thank you so much!