Meta's SAM3 open source vision model tracks objects reliably
Original: there's no catch; SAM3 is open source and really good
Source: x.com ↗
Who: Posted by @SkalskiP (computer vision practitioner and educator, known for hands-on ML tutorials), responding to a question from @ludwigABAP. The underlying technology is from .
What's new: SkalskiP pushes back on the idea that models are underused by arguing there is no hidden catch: SAM3 is genuinely free, open source, and performs at a high level on hard real-world tasks like tracking basketball players through a crowded, fast-moving scene.
Why it matters: Most powerful vision tools from big companies come with usage fees or licensing restrictions. SAM3 being fully open source means any developer can build products on top of it without paying or asking permission. SkalskiP calls it his favorite model ever, which is a strong signal from someone who works in the space daily.
Caveats: The post is a brief opinion on social media, not a technical review. No benchmark numbers, comparisons to competing tools, or failure cases are cited. The original question from @ludwigABAP about why adoption seems low remains only partially answered: enthusiasm from practitioners does not explain the apparent gap between capability and commercial product use.