sam-3

Digitising Scrabble boards with SAM 3 and Perception Encoder.

Apprentice is a web app that learns a hands-on procedure from one video, then checks later recordings for skipped, out of order or steps which could be incorrect. Built with Meta’s Muse Spark and SAM 3.1: models propose, pixels ground, rules decide.

Measures how long a brand is actually on screen in creator video and checks it against the contract. Built with Meta SAM 3.1, Muse Spark and Muse Voice Transcribe.

Film yourself painting. Step out. The painting stays. Every pixel from the camera, nothing invented. SAM 3.1 and Muse Spark, Meta Global AI Developer Hackathon 2026.

Drone search and rescue for a volunteer with one drone: where to fly next, and whether the missing person is in the photos. Built on Muse Spark, SAM 3.1 and Muse Voice Transcribe.

Packing-evidence assistant for Indonesian online sellers, built on Meta Muse Spark 1.3 and SAM 3.1. Successor to ParcelProof.

SceneProof — counterfactual world-model verifier for robotics & embodied AI. Segment it, intervene, prove it. Built for the Meta Global AI Developer Hackathon 2026.

Photograph a loaded trailer; LoadMaster checks it against the federal cargo securement rules (49 CFR 393) with SAM 3.1 + a deterministic rules engine + a Muse Spark second opinion, and says in plain words what to fix.

Optical visual referee for AI coding agents. Dual-instrument verification (Meta SAM 3.1 + Pixel NCC) calibrated to a 3σ noise floor. CLI, MCP & Workbench.

Research framework for zero-shot anomaly detection and segmentation on the WS2 dataset using vision-language models, grounding methods, and SAM-based segmentation pipelines.

QorTime AI: Meta Muse Spark, Muse Wonder and SAM 3.1 explain, red-team and visualize a deterministic video-game integrity referee (anti-cheat + verified skill, signed match receipts). The AI never decides a verdict. Meta Global AI Developer Hackathon, Indie Developer track.

Privacy-preserving SAM3 distillation pipeline for fast image/video redaction and inpainting, with optional offline LLM prompt expansion.

Tennis movement data from a coach's training video with Meta SAM 3.1 and Muse Spark 1.3 (Meta Global AI Developer Hackathon)

Pinned - one link, one game, hosted by an AI. Meta Global AI Developer Hackathon (Indie, Build for Your World).

Predicting a promptable video tracker's zero-shot reliability on unseen species and places - detection and association separately - from label-free distances. MSc dissertation.

Website for CourtSense: tennis movement data from one camera with Meta SAM 3.1 and Muse Spark (Meta Global AI Developer Hackathon)

FrontierSight AI — From video to verifiable physical-world intelligence. Meta Global AI Developer Hackathon 2026 (Startup Track, Push the Frontier).