Notes on research, 3D vision, generative models, and things (economics, biking, life, coffee) I've been reading and thinking about.
-
15 September 2026
From Videos to Pointmaps to Simulated Videos: DA3 + cameras over time, Gaussians, and video models along new VLA trajectories
DA3 with the calibrated rig stitched over time into one metric world, Gaussians rendered along trajectories the car never drove (sampled by Alpamayo-R1), and 3D-informed video models (GEN3C, Vista4D) filling what the rig never saw, then reconstructed again.
-
24 August 2026
Feed-Forward 3D Reconstruction Meets a Real AV Camera Rig
Testing DA3, VGGT and friends on NVIDIA's 7-camera AV dataset: camera conditioning fixes the geometry, classical calibration fixes the meter, and the leftover error is a training-data story.
-
24 August 2026
Hello, world
Why I'm starting to write here, and what to expect.