I was exploring the capabilities of the blueprint. I have some questions regarding some customisations:
-
Custom Models: I see the ingestion pipeline uses Grounding Dino for detection, custom reid module for tracking and SAM for segmentation. Can i use my own custom models in place of those, e.g. Yolo for detection, my own custom tracker or yolo for segmentation. If yes, does it need a code change or only config change
-
Custom Pipeline: If say i want to only use the ingestion pipeline, is it possible to disable the retrieval pipeline and run only the ingestion pipeline till CV and tracking pipeline.
-
Deepstream Usage: Does the ingestion pipeline use deepstream underneath?
-
Business Logic: Say if i want to implement some business logic on the output of the CV and tracking pipeline, how can i do it?
-
Streams Support: Is there any benchmark on the number of streams that can be supported for say a given type of system, e.g. (20 RTSP streams for A100)
-
Multiple Models: Can i use multiple independent models in my CV and tracking pipeline for 1 RTSP stream, e.g. (Detection + Segmentation or Detection (type A) + Detection (type B) ). If yes, what would be the best way to aggregate there outputs.
-
Action Recognition: Can i use something like MViT (Multi scale vision transformer ) models as one of the models in my ingestion pipeline (CV and tracking)