Vision Transformer or other Temporal Vision Models

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Please provide complete information as applicable to your setup.

**• Orin AGX 64GB
**• 7 or 6.2

Hi!

I’m wondering what the best way would be to have spatial-temporal models, such as RNNs or vision transformers that can aggregate information over a long(er) time duration inside Deepstream? I want to go away from static-image object detectors. I know of this, but it seems outdated and I’m not sure how easy it is to adapt: DeepStream 3D Action Recognition App — DeepStream 6.2 Release documentation (nvidia.com)

Is there any reference or such?

Thanks!

3

The DS 3D action sample is available in the latest DeepStream 7.0.

4

Thank you, does that mean this is the only example available in the direction of RNNs/Temporal models/…?