What is a rose🌹visually? A rose is characterized by its intrinsics: the distribution of geometry, texture, material.
We learn a generative model that discovers object intrinsics from just a few instances in a single image - a task humans can do with ease.
#CVPR23 poster Tue AM

How are Deep Neural Networks black-boxes if you can visualize them in an 'exact' manner?

Our new #CVPR23 paper, presents a fast and scalable PyTorch toolbox to visualize the linear regions, aka partition+decision boundary, of any DNN (red🔻)!
bit.ly/splinecam

🧵 1/N

This is the big one! Repealing the social media ban.

Jon Barron, Michael Black, Dmytro Mishkin 🇺🇦, and others speaking in favor

#CVPR 23 #CVPR #CVPR2024

Getting ready for the demo at #CVPR23 @MetaAI booth tomorrow.

I'm thrilled to announce that two of my papers have been accepted at #CVPR23 ! Huge thanks to my amazing intern manager Shunsuke Saito and collaborators for their hard work and dedication. See project website for more details: junxuan-li.github.io/megane and stay tuned for more information!

🎉 Thrilled to share that our work on computationally budgeted continual learning has been accepted to #CVPR23 !
📝 Check out our paper “Computationally Budgeted Continual Learning: What Does Matter?” arxiv.org/abs/2303.11165
🧐 In this 🧵 we delve into the key findings of our work

Track #HumanObjectInteractions from a video in 3D. Transformer based VisTracker reasons about occlusions and leverages surrounding frames to make correct predictions under occlusion. virtualhumans.mpi-inf.mpg.de/VisTracker
Poster: Tuesday, 1630PM, West Exhibit Nr. 59 #CVPR23 #3DHumans

Visual odometry is a basic function for embodied AI. At #CVPR23 we will present a multi-modal & modality-invariant visual odometry framework called Visual Odometry Transformer (VOT). Also I give a talk on multi-modal learning on several projects at the Multiearth w/ on Mon.
🧵

中国の信号で自動運転するの難しすぎるだろ😂
#CVPR23

Excited to announce our new benchmark for Online Continual Learning (OCL) with real-world settings, which has been accepted as a 'Highlight' in #CVPR23 .

“Real-Time Evaluation in Online Continual Learning: A New Hope”. arxiv.org/abs/2302.01047

Let's take a closer look!🧵
1/7

FAIR and ⁦Georgia Tech Computing⁩ researchers demoing our sim2real work on ⁦Boston Dynamics⁩ Spot at ⁦@MetaAI⁩ booth at #CVPR23 .

Coming to #CVPR23 ? Come to my talk on Sunday morning at the GAZE workshop gazeworkshop.github.io/2023/ to hear about Synthetic Data for Gaze Estimation & More. Find out what worked 🙌🏻, what didn't 🤔, and what's next 🚀.

We are happy to announce that our paper 'MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation' was accepted at #CVPR23 !
By enforcing consistency of predictions from masked images, MIC enhances the adaptation to an unlabeled target domain. github.com/lhoyer/MIC

Amazing presentation at CVPR23 by Phil Duan gave.

The advancements happening every 6 months are astounding, resulting in better outcomes with less compute on the FSD chip.

Soon, Tesla isn’t going to be limited by compute power on the training side either. I wouldn’t bet…

Yue is presenting his very popular LaViLa work (facebookresearch.github.io/LaViLa/) from his internship at Meta at the #CVPR23 poster session that “starts” in 15 minutes! Come by if you’re interested in LLMs and videos!

#CVPR23 is over, but the memories and social bounds it created will last forever.

I want to deeply thank all the people that spent time with me and made this such an amazing experience.

CVPR is what really pays off the hard work of a whole year.

See you in Paris for ICCV!

First time attending LatinX in AI (LXAI) workshop at #CVPR23 in person!

I got a bit carried away in my oral presentation, but feel free to ask any question during the poster session.

📢Panoptic Lifting for 3D Scene Understanding with Neural Fields #CVPR23 highlight!

Given only posed RGB images of a scene, we optimize a panoptic radiance field representing color, depth, semantics, and instances at any point in space.

Vid: youtu.be/QtsiL-6rSuM
Yawar Siddiqui

#CVPR23 is featuring ~2400 papers over 3 days (or 6 sessions). In other words, ~400 posters/session!

It remains to be seen if in the future a 3 day conference will be enough to ingest (even at a glance) an insane amount of information.

So CVPR23 is now over, and there were so many papers on NeRF. This conference alone had 175 NeRF-related ones. So this year, I built a tool to generate short descriptions for my blog post and handle the formatting for me.

markboss.me/post/nerf_at_c…