Ran Meta's SAM 3 on a chase sequence from Uncharted 4 to test its performance on video game footage. It did pretty well. Simple additional prompts are enough to extend detection to many other objects in the scene.

You can find more details about Meta's SAM3 in this post.

I don't see this making aimbotting worse in online games right now because running a model like this continuously is still pretty resource heavy. As models get more efficient though, it could become a problem down the line.

SAM3 shows comparable performance on both real world and video game footage. This kind of consistency is especially useful for robotics, where training often starts in simulated environments and only small adjustments are needed when deploying in the real world.

is this real time? the signal I'm getting here is I can legit just use SAM 3 for in-game features

SAM3 + YOLO 👀

游戏拟真感拉满

wild potential

clean run tbh

wild if it nails fog

Oh shit just had an idea that game companies will not like……

稳得一批

This project seems solid, reach me 📈

稳了

crazy that it handled motion blur that wrll

Real world robotics training gonnq be way easier if SAM3 keeps this level of consistency

实景怕是更狠

real world sim bridge

稳了

稳得有点吓人

New technology opens up endless possibilities for gaming.

could be huge for sim devs

The best chase sequence in gaming btw

Nice test bro 🫡

SAM3 crushing it on Uncharted 4 already? Wild. Games about to get next level segmented fr (34 chars)

Object detection accuracy improved significantly during the Uncharted 4 chase scene analysis.

Uncharted chase + SAM3 = cinema vibes

游戏里这么稳？那实景训练直接开冲

SAM 3 passed chase sequence test with flying colors.

çılgın şey

game studios gonna hate this