lukehollis

CarveAnything takes an input monocular image and generates a 3d model with predictive volumetric content, physics properties, and destruction mechanics for simulating cutting, lighting, and staging objects--so it may be adapted to the Halloween season.

One can use this to automate robotics policy evaluation and training on diverse objects, and more on methods is in the paper. We use a combination of semantic features and predictive per-Gaussian physics to transform individual complex objects or entire scenes to simulation environments in Isaac Lab or MuJoCo. Pumpkin carving itself is a difficult task to model accurately in these simulators.

Happy carving, HN! Glad to know of failure modes with any generated objects/carves.

Text to 3d simulation on a map. It does historical or fictitious events pretty well if it's interesting: https://mused.com/map/

I was working on world models / generative environments but without the training data available as an independent researcher, ended up focusing on building with existing geospatial data.

The same architecture of the '24 Genie paper's dynamics model is instead trained on historical data for risk analysis and creating a heatmap in the 2d map. I'll try to adapt this for a more generalizable urban mobility model as well.

Thanks!

you're the best of us. How'd it do? Can you share it?

Ayvri is impressive, and thanks for the feedback - I hear what you're saying about the convincing replica

Updated -- the geocoding was moving the map around too quickly

Thanks, this should be updated for a better landing on mobile

Thanks! The worst is the elevation API -- but yes, I monitored costs throughout today, it's not prohibitively expensive.

Okay back online in 3d

Okay back online in 3d too!

Thanks!

Here's a video -- the 3d map tiles quota was hit https://www.youtube.com/watch?v=zXS9sNcDLJU

Should be back shortly!

Sorry, hit a quota on the 3d map tiles - here's video https://www.youtube.com/watch?v=zXS9sNcDLJU

And requested increase, should be back soon!

Hey defaulting to 2D map temporarily! 3d will work shortly

This is the best thing that anyone has simulated yet hahah

3d version exhausted quota -- will be back shortly! Video of the 3d version https://www.youtube.com/watch?v=zXS9sNcDLJU

Sorry, exhausted quota on the 3d tiles from Google Maps Platform -- should have workaround up shortly

Asking for a quota increase with support! Will build in bring-your-own-key

Update: sorry google maps 3d tiles aren't displaying!

Video: https://www.youtube.com/watch?v=zXS9sNcDLJU

The Google Maps 3d Photorealistic Tiles are returning 429 Error too many requests Looks like will need to build in the ability for people to bring their own Google Maps API key. Fix up shortly

Wow, thanks for writing -- would love to see what you're building when you have a link to share! My emails in my profile or just will keep an eye out in general for this workflow.

It was specifically the time in the tunnels from 6:00 AM - noon every day that was an out-of-the-ordinary time and energy investment. It was necessary to be there, so I evaluated as "do things at dont scale."

Thanks for letting me know! That makes good sense, and I did this in other tours. I'll see what I can do.

Thanks, I always wonder if it'll be too slow for people to follow the whole pathway--this will definitely be fixed in the desktop version.

Thanks for the correction!

Thanks so much! Does it look very different / did you go in the tunnels?

Yes, these scans also work in WebXR using Matterport's default VR viewer -- I have a basic page setup at https://mused.com/vr/ -- but I'll get the Copan tours added there.

For the majority of the monuments that I worked on, they were ritually destroyed, buried, and built on top of -- with the Rosalila temple being the main exception. The tombs typically have some type of arch and the temple area and staircases are kind of built around or on top of them.

Yes, I 3d captured the murals at San Bartolo, and they're somewhere in the pipeline of being ready to be made public soon.

Very cool.. bravo -- it looks so good

Ha! that's the best theory yet