MemoryWorld
We’re building MemoryWorld, a new way to capture experiences, using the insane power of Argon to reconstruct and iteratively refine 3D worlds from your photos and geodata.
Core trick
Reconstructs 3D scenes from photos and geodata, then refines them in a render-compare loop with a model.
Tags
Description
We’re building MemoryWorld, a new way to capture experiences, using the insane power of Argon to reconstruct and iteratively refine 3D worlds from your photos and geodata.
README excerpt
# Stanford: a memory world <a href="media/showcase.mp4"><img src="media/showcase.gif" alt="Atlas showcase: posing the photo with Argon 4, adding USGS terrain context, fusing it all into one memory world" width="100%"></a> <p align="center"><a href="media/showcase.mp4">▶ Watch the full-quality video (15 s)</a></p> Photos turned into places you can fly into, at Stanford: beach volleyball at sunset (23 June 2026, 19:11), the Main Quad on a bike (12 October, 17:19), a picnic breakfast at Meyer Green (13 October, 08:36) and a night on a bench with friends. The journey starts above California, descends onto each spot, lands in its photo, and hops across campus (a few hundred metres up) to the next, then rises back out. ## Run it ```sh python3 serve.py # then open http://localhost:8771 ``` Any static web server works; `serve.py` only adds the right type for `.ply` files and disables caching. | Input | What it does | | --- | --- | | scroll, ← → | move through the journey | | space | drift on | | drag | look around | | W A S D | wander | | 1–4 | fly to a memory (volleyball, Main Quad, Meyer Green, the bench) | | M | back up to the map | ## How it was made - **Place:** the photo's GPS point, USGS 3DEP elevation and USGS NAIP imagery around it (public domain), and the California map from the same sources. - **Descent:** ten views from 450 m down to eye level, generated with World Labs Atlas from the photo and the aerial imagery around it (relit to the morning sun