Hi,
I’m working on DLStreamer at Intel - a GStreamer plugin for AI inference using OpenVINO. I wanted to share our experience migrating to GstAnalytics and get feedback from the community.
We are currently transitioning from legacy metadata (GstVideoRegionOfInterestMeta + custom structures) to the GstAnalytics framework.
Our implementation status:
Currently supported:
-
GstAnalyticsRelationMeta
-
GstAnalyticsODMtd
-
GstAnalyticsClsMtd
-
GstAnalyticsTrackingMtd
-
GstAnalyticsGroupMtd and GstAnalyticsKeypointMtd (integrated from upstream before official release)
Planned:
-
GstAnalyticsTensorMtd
-
GstAnalyticsSegmentationMtd
We are running a hybrid approach during migration - both legacy and GstAnalytics metadata coexist, with GstAnalytics taking priority where implemented.
Questions for the community:
Is anyone here using GstAnalytics for AI inference?
Are there best practices for the relationships between specific GstAnalytics instances? Is this standardised in any way?
Our current metadata implementation details are documented here if anyone is interested in the technical approach: https://docs.openedgeplatform.intel.com/dev/edge-ai-libraries/dlstreamer/dev_guide/metadata.html
Looking forward to contributing back any insights from our migration experience.