Brain Builder » Segmentation » Best Practices: Video Annotation

On This Page:

Best Practices: Video Annotation

Lifelong-AI Annotation

With any tool, there are situations in which it works well (e.g. hammer + nail) and situations when it does not (e.g. hammer + screw). Some of Brain Builder's tools use Neurala's patented Lifelong-AI to enhance the data tagging workflow. In order to make the most of these tools, it's important you understand when they work best.

Before you begin...

If you haven't already done so, check out the documentation pages for Tagging Data and Video Tagging

Warning Always be sure to use the latest version of the Chrome Browser with Brain Builder, as it is currently the only browser that Brain Builder officially supports.

AI Video Annotator

Neurala's AI Video Annotator learns when you tag one frame and applies that tag to subsequent video frames going forward.

Data Types & Tips

  • Brain Builder supports the following file types: .mp4 .m4v .avi .wmv .mov (.MP4, .M4V, .AVI, .MOV, .WMV)

  • Brain Builder supports a maximum upload size of 16,000 pixels tall and wide

  • We recommend using videos that are at a standard frame rate of 30 FPS. The AI Annotator works best when there is smooth motion from one frame to the next. Low frame-rate videos often skip frames, resulting in choppier motion from frame to frame, which results in poorer AI-assisted tagging.

Works Best On...

The Video Annotator works best with objects that:

  • Are visually distinct from the background
  • Are in focus
  • Are relatively prominent in the frame
  • Maintain a similar appearance from one frame to the next

Does Not Work Best On...

The AI Video Annotator delivers worse results on:

  • Very small objects
  • Low-resolution objects or videos
  • Low-visibility objects (i.e. dark scenes, camouflaged objects)
  • Overlapping objects of separate classes
  • Fast-moving, blurry objects
  • Objects that change significantly in appearance from frame to frame (e.g. tagging a chameleon that changes from green to blue)

Video Scene Changes

The AI Annotator looks at overall continuity from the first frame of annotation to generate predictions, so each area where the object of interest looks distinct throughout the video should be worked on separately. This includes significant changes in view, perspective, or background. For each section, tag the first frame, define the start and end points, and run the Annotation.

Next: Best Practices: Smart Fill & Snap-to-Fit >>