Hacker News
X / Twitter
the design software we use for these lots
Human3R: Everyone Everywhere All at Once
Note: I recorded the video from the interactive demo on their project page (linked in the comment below).
Abstract (excerpt):
Human3R jointly recovers global multi-person SMPL-X bodies ("everyone"), dense 3D scenes ("everywhere"), and camera trajectories in a single forward pass ("all-at-once").
Our method builds upon the 4D online reconstruction model CUT3R and uses parameter-efficient visual prompt tuning to preserve CUT3R's rich spatiotemporal priors while enabling direct readout of multiple SMPL-X bodies.
Human3R is a unified model that eliminates heavy dependencies and iterative refinement. After being trained on the relatively small-scale synthetic dataset BEDLAM for just one day on one GPU, it achieves superior performance with remarkable efficiency: it reconstructs multiple humans in a one-shot manner, along with 3D scenes, in one stage, at real-time speed (15 FPS) with a low memory footprint (8 GB).
Huge miss by Adobe not relaunching Dreamweaver
📽️ Check out Visual Odometry Transformer! VoT is an end-to-end model for getting accurate metric camera poses from monocular videos.
https://t.co/6tVXVt6mTx
VoT does not require camera calibration parameters, post-optimization, and operates in real-time, capable of processing thousands of frames. It is trained on a vast amount of real-world indoor data, but can work just fine in outdoor scenarios. It uses only camera poses as supervision - no optical flow, intrinsics, point clouds, or tracks - making it broadly accessible.
We experimented with different backbones, camera pose representations, scalability, and attention mechanisms. Our evaluation spans hundreds of full-length videos across various metrics, without aligning the predicted trajectory to the ground truth, to simulate a real-world application.
Thanks to the team, @kienduynguyen94, @theogevers, @cgmsnoek, and @Martin_R_Oswald from the @UvA_Amsterdam!