Video-level classification

[ehofesmann]

My actions before raising this issue

• Read/searched the docs
• Searched past issues

Currently, tags can only be applied to video frames. It would be useful to provide a way to classify an entire video with a single tag.

Expected Behaviour

Add either a new label type or an option to tags allowing them to be defined over an entire video. Or even better, allow tags to be defined over a frame range which would also allow for temporal clip annotation.

Current Behaviour

Currently, a workaround is to just tag the first frame, then post-process the labels to upgrade the frame-level tag to a video-level tag. This is not ideal since it relies on conventions that can accidentally be broken by annotators.

Context

Some of the users of the FiftyOne and CVAT integration are working heavily with video data. They perform the majority of their annotation in CVAT, but due to the lack of a way to annotate a whole video with a single tag, they have been looking into other annotation tools for this purpose. Given that this seems like a relatively simple label type, it may be worth adding to CVAT natively.

[bsekachev]

Maybe related #1228

[ehofesmann]

@bsekachev Ah yeah, that is similar! What I'm specifically looking for is not an attribute on an object annotation, but just a video-level attribute directly that is immutable and cannot be modified using keyframes.

[nmanovic]

It is connected with #5003