Investigative video editor - powered by Meta SAM2

Blur, track, and modify objects in video for investigation

Lawyers sometimes need to edit video evidence. Common use cases include:

  • Redacting sensitive information: Blurring faces, license plates, or confidential details to protect privacy.
  • Clipping relevant segments: Extracting only the portions of video that are relevant to the case.
  • Annotating or highlighting: Adding timestamps, captions, or arrows to clarify events for the court.
  • Synchronizing with other evidence: Aligning video with audio, transcripts, or other exhibits for clear presentation.

For instance, shown below is a sample video where we want to track three people and blur out everything else. The original video looks like this:

Blurring everything except the three people in this video is a lot of work. The aim of this project is to automate this. (Video by Kelly from Pexels).

These tasks are typically unavoidable and soul-crushing parts of the job. But with the arrival of new video-centric models, like Meta’s Segment Anything 2 (SAM2), this has changed.

With SAM2, it is now possible to reliably track objects and people in a video.

Results

Suppose we want to highlight a group of three people for a presentation.

Example initial frame with the selection of three highlighted people.

The tool allows automatic tracking of selected objects and is robust against occlusion (occlusion is when an object is temporarily hidden from sight). For instance, the above airport scene, in a few simple clicks, can become this:

We track and highlight the only three persons of interest, with SAM2.

Here’s another pair of videos where there’s even more significant occlusion, with an even longer tracking time. The original:

Video by CityXcape from Pexels

With just a few clicks, we select three guards and proceed with tracking. We remove the video background to leave only the guards in place:

Results of the auto-background removal feature, with SAM2.

Using this video editing tool is easy. Selecting a face and tracking it is done with just a few mouse clicks. A button allows for the selection of all things of interest. For the above video, we only needed 12 clicks (marked on the timeline as blue, orange, and cyan dots) to select the three guards. The interface allows easy visualization and correction of the marks.

Example of the graphical application that allows tracking objects inside a video—in this case, we needed only 12 clicks for the three guards.