Introducing MegaSaM! 🎥 Accurate, fast, & robust structure + camera estimation from casual monocular videos of dynamic scenes! MegaSaM outputs camera parameters and consistent video depth, scaling to long videos with unconstrained camera paths and complex scene dynamics!

We ran DUSt3R on our cartoon reconstruction setting and found that it struggles (even with ground truth correspondences)! Our proposed work is designed to handle this challenging setting where input images are hand-drawn and non-geometrically consistent. toon3d.studio

I'm not at neurips, but David McAllister and Aleksander Holynski are just about to start their poster session on our paper that presents a unified framework for explaining SDS and its variants sds-bridge.github.io

Big congrats to the Luma AI team!!!

Code and model out for Cut3r✨, including the meta-data/processing code to reproduce *all data* we used for training the model!!!!

Don't have the massive compute clusters for training a big diffusion model? 💸 Our work lets you train on smaller GPU clusters, then combine them later—no loss in performance; in fact, it performs better! A step toward more accessible diffusion models ➡️ decentralizeddiffusion.github.io

Ethan's new work pushing on completing a big 3D scene from casual capture. Still a very hard problem since you need to generate new content while integrating a lot of context. This multi-view diffusion model is opensourced for the community to make more progress!

Synergy in reasoning about humans, cameras, and places: Adding people to SfM helps all around🌏 Cameras ground people, humans provide correspondences & scale, and structure keeps it stable. We call it HSfM! Demo code out now: github.com/hongsukchoi/HS…

Creat videos with arbitrary explicit camera control! Looks really consistent! It’s hang’s last PhD work and he is also in the market 😄!!

We added some buttons on the top of the Toon3D Labeler labeler.toon3d.studio so you can explore the Toon3D Dataset toon3d.studio. Beyond 3D reconstruction of hand-drawn cartoons, I think this dataset might be useful as a benchmark for 2D correspondence methods.

PyRoki: new open-source toolbox from our lab. Makes robot kinematics optimization easy with python—flexible autodiff with (my favorite part) a Levenberg-Marquardt solver that properly handles rigid 3D transforms via Lie groups + an awesome visualizer 🔥 github.com/chungmin99/pyr…

Teaching humanoids how to move through the world from video. videomimic.net