[DOCS] Is there a documentation for converting video dataset into frames or images and vice versa?

[DWSuryo]

URL(s) with the issue

https://docs.voxel51.com/recipes/convert_datasets.html

Description of proposal (what needs changing)

Title. As the link I refer, there are some things I want to talk. I think this documentation only covers the dataset conversion for the same media type, for example, image to image. With that, I want to ask:

1. Is there a documentation or at least a tutorial to convert the video to image dataset and vice versa?
2. At that link, looks like the notebook heavily uses CLI commands. Is there a python friendly tutorial that reduces CLI usage?

That is all I can say for now. I will update later.

Willingness to contribute

The FiftyOne Community encourages documentation contributions. Would you or another member of your organization be willing to contribute a fix for this documentation issue to the FiftyOne codebase?

• Yes. I can contribute a documentation fix independently
• Yes. I would be willing to contribute a documentation fix with guidance from the FiftyOne community
• No. I cannot contribute a documentation fix at this time

[DWSuryo]

Update: I've tested myself a code for extracting video to frame.

Here's the code: Copy_of_FiftyOne_quickstart_video.ipynb

First thing I can do is just converting video to video dataset. Not sure for video to image dataset.

It's still rough and barebones though. The target is to make a label for each frame. I think I need help for this one.

Update (Jan 25, 2025): I've updated the code as suggested by using clone().to_frames to duplicate dataset and convert into frames at once

[swheaton]

I think you are looking for to_frames() ! This will give you a view into your dataset. You could save this view to easily pull it up again, or if you want a separate dataset, clone() the view into a new dataset.

https://docs.voxel51.com/user_guide/using_views.html#frame-views

[DWSuryo]

Nice! I overlooked this one as I came back to use the tutorial I'd mentioned before. So, in updated code (Jan 25, 2025), I've updated that it can convert into COCO and YOLOv5 (Ultralyticss) dataset. If I recall correctly, COCO uses one json for all annotations whereas Ultralytics separates the annotation per image. Am I correct at this one?