ICESight is a photo-based tool that uses AI to handle moderation and verification, with human users acting as verifiers. When someone uploads a photo of an ICE sighting, it runs through an object detection model to find possible agents or vehicles. Each detected area is then sent to another AI service that checks if the scene looks authentic.
If it passes those checks, the system saves visual embeddings for each agent in a database and compares them against future uploads. Verified photos are added to a public map where other users can review and confirm sightings.
I hope this can become a useful public dataset that helps improve transparency and public safety. Feedback very welcome!