Crumb
c/localllama · by thickertoofan@lemm.ee · 1y

SpatialLM, a 1B model capable of spatial identification, using 3d point cloud data. The video demo is amazing.

https://manycore-research.github.io/SpatialLM/

26 points · 5 comments · view on lemmy.world

5 Comments

Tar_alcaran@sh.itjust.works · 10 pts · 1y (1 reply)

The demo looks pretty impressive, but it's a prerecorded demo we know nothing about. SO many AI companies have been lying about their benchmarks.

pennomi@lemmy.world · 1 pts · 1y
[ removed ]
Karkitoo@lemmy.ml · 5 pts · 1y (1 reply)

Looks impressive and it's truly open-source.

However, I see it requires CUDA. Could it run anyway:

  1. Without this?
  2. With AMD hardware?
  3. On mobile (as the model is only 1B) ?
thickertoofan@lemm.ee · 2 pts · 1y

I think the bigger bottleneck is SLAM, running that is intensive, it wont directly run on video, and SLAM is tough i guess, reading the repo doesn't give any clues of it being able to run on CPU inference.

hendrik@palaver.p3x.de · 3 pts · 1y

Now build a robot / autonomous (remote controlled) car / robo vacuum with that.

Aceticon@lemmy.dbzer0.com · 3 pts · 1y

Isn't that basically a variant of photogrametry but enhanced by AI so as to need fewer photos?