Annotate Videos
Annotate video frame sequences with freeform text to train a vision-language model on temporal and visual reasoning tasks.
Datature Vi supports freeform text annotations for videos. Unlike image annotation, video annotation works with frame sequences: you select a range of frames using the video scrubber, then write text annotations that describe what happens across those frames.
A dataset with uploaded videos. Create a dataset if you don't have one yet.
Annotation types
Freeform text
Freeform text annotation for videos lets you select a sequence of frames using the timeline scrubber, then write any structured or unstructured text for that sequence. This is ideal for describing actions, events, physics behaviors, and temporal relationships that span multiple frames.
The result: a model trained to understand and reason about video content over time.
Typical use cases:
- Action recognition and description
- Temporal event analysis
- Physics and behavior rule verification
- Video captioning and scene understanding
- Activity monitoring and compliance checking
Annotate videos with freeform text
How video annotation differs from image annotation
Video annotation adds a temporal dimension. Instead of annotating a single static image, you work with frame sequences on a timeline.
Annotation workflow
- Upload videos to your dataset
- Open the annotator from your dataset's Annotate tab
- Select a video from the thumbnail strip
- Use the timeline scrubber to select frame sequences
- Write freeform text annotations for each sequence
- Review coverage using the dataset overview
- Train your model using the annotated dataset
Next steps
Updated 6 days ago