Name: VideoGPT+: integrating dual encoding for enhanced video understanding
Uploaded: 2024-07-25
Channel: MBZUAI
Description: VideoGPT+, a video conversation model built at the university. A "joint dual encoding mechanism" reads the video both frame by frame and as a whole so the model can reason over space and over time, with a semi automated pipeline that extends the instruction tuning data and a new benchmark covering video types that range from "sports videos to movie scenes".

Back

Statement 1 / ~3:33One statement was said in this videoOne statement

Provenance

Speaker: Researcher presenting the model (name inferred)
Entity: MBZUAI
Sector: Media and content
Date: 25 Jul 2024
Said at: 3:33, estimated from the transcript, may drift by up to a minute
Language: English
Source video: VideoGPT+: integrating dual encoding for enhanced video understanding
Channel: MBZUAI
Original link: Open on YouTube