← sources

Video events grounding and captioning ?

source: reddit (reddit:StableDiffusion)published 2026-09-26 02:40ingested 2026-09-27 02:38

retained text

Multiple CCTVs camers, I have solved the Identity and tracking across camera, now need a model that converts these into 

[00-02] Abhishek exits the classroom. 
[02-06] Arjun gets up and walks to the
back of the classroom. 
[11-14] Arjun walks forward to Mahesh and Darsh's desk.
[14-18] Arjun interacts with Mahesh and Darsh
[19-23] Arjun walks back towards
the rear of the classroom. 
[34-37] Arjun sits down at a desk in the back.

Timestamps from the video not necessary
Maybe achieve this first, then hopefully make it realtime 

40 camers and 5060ti is what I have 

document fetched 2026-09-27 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1wqf5oi/video_events_grounding_and_captioning