Recipes for videos

[Iran R. Roman]

Given that each video is unscripted, it would be challenging to indicate specific recipes that were carried out. However, have you considered including the recipe(s) carried out in each video as another annotation?

This would be useful to:

• assess video summarization methods
• supervise model attention to actions related to a specific "goal" while ignoring other actions
  • Example: cooking the rice for a meal VS washing the dishes to maximize use of time while waiting for the rice

1. Do you have metadata pertaining to specific recipes carried out in each video?
   • The language of these recipes would be contrastingly different compared to the granular description of all actions carried out in a video.
2. Did you try to generate such metadata using existing models derived from other datasets (like the Recipe1M+)?
3. How would you go about annotating such ground-truth metadata for all or some of the videos with existing models/methods/metadata?

[Dima Damen]

Thanks for your suggestion. This is not something we considered or looked into.

Usually these GitHub issues are opened for concerns with the annotations, so I'll close this for now as I do not believe it's related to a concern with the annotations. If you have ideas/suggestions please email the project mailbox instead.

[Iran R. Roman]

@dimadamen. Thank you very much for the clarification.