You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
One way LLMs can do video processing is via image segmentation, with tokens as labels for each entity in ascii.
We can really reduce context length required if doing this method.
Adding this recent repo for pre-processing, could then be used for next frame prediction, and with action in the loop (as demonstrated previously at Mistral SF hackathon): https://github.com/facebookresearch/segment-anything-2
The text was updated successfully, but these errors were encountered:
One way LLMs can do video processing is via image segmentation, with tokens as labels for each entity in ascii.
We can really reduce context length required if doing this method.
Adding this recent repo for pre-processing, could then be used for next frame prediction, and with action in the loop (as demonstrated previously at Mistral SF hackathon):
https://github.com/facebookresearch/segment-anything-2
The text was updated successfully, but these errors were encountered: