🥬 Rotational neural keypoint 🥬

👉RaCo is a rotationally equivariant neural keypoint detector that identifies stable and repeatable keypoints. Code under Apache 2.0💙

👉Review https://lnkd.in/p/ezQzUwju
👉Paper https://arxiv.org/pdf/2602.15755
👉Repo https://github.com/cvg/RaCo

Hi everyone 👋

over the past months, engagement on my posts has grown a lot, which tells me the content you value is getting better. At the same time, impressions have dropped, and I think a big part of the reason is the flood of low-quality viral posts filling many of our feeds. PROVE ME WRONG 😂

The best way to push back is to engage with the posts that are actually worth your time. Open the post on LinkedIn, then leave a comment (even a couple of lines with your take), and like, share, or save it if it's useful. Comments are the most valuable, but every action helps. With ~100,000 of us following along, this is how we tell the algorithm, and me, what deserves more space.

Fight fire with fire!

🐇Segment where you look🐇

👉KAUST (+Adobe) revisited the interface through which a vision-language model passes the location of its referent to a segmentation decoder. Code announced💙

👉Review https://lnkd.in/p/er_ifxGR
👉Paper https://lnkd.in/eATXw98R
👉Project https://lnkd.in/eXSWvxPN
👉Repo TBA

🧊SOTA category-level 9D Pose🧊

👉SJTU at #NeurIPS 2026 with the new SOTA among RGB-based methods for category-level 9D pose estimation. Code announced💙

👉Review https://lnkd.in/p/e5BvM7Ua
👉Paper https://arxiv.org/pdf/2610.18374
👉Project https://souglad21.github.io/SchemaPose/
👉Repo TBA

🎊 Pixel-Level 4D Humans 🎊

👉Dense4D from the Imperial College is the first feed-forward model dedicated to dense 4D correspondence across the body, clothing and hair from a set of monocular RGB images, without a body template. Code announced💙

👉Review https://lnkd.in/p/e9GXGggz
👉Paper https://arxiv.org/pdf/2610.22325
👉Project https://alpguler.com/dense4d/
👉Repo TBA

Back to post on Twitter 😈

https://x.com/ai_withpapers

🔥700 FPS Humans Recovery🔥

👉One model, any sparse signal: OmniPoser recovers full-body 3D human motion from 3D keypoints, sparse IMU rotations, head-mounted 6-DoF poses, or any subset or combination of these signals. All within a SINGLE model. Repo under nc-cc license💙

👉Review https://lnkd.in/p/et-Pk-Dy
👉Paper https://arxiv.org/pdf/2610.19857
👉Project https://jamesbridgewater51.github.io/omniposer-eccv/
👉Repo https://github.com/JamesBridgewater51/omniposer

🏵️Lang-Informed Lossless Encoder🏵️

👉Stanford's LILIE: an invertible image transform trained such that a vision-language model can recover captions from noised features. Code, encoders, and generative models💙

👉Review https://lnkd.in/p/dq3SK_nc
👉Paper https://arxiv.org/pdf/2610.23653
👉Project https://www.suchiragarwal.com/lilie/
👉Repo https://github.com/SuchirAgarwal/lilie

🔥SOTA Video Matting🔥

👉VideoMaMa++ (#NeurIPS 2026): mask-guided video matting framework that recovers fine-grained details (e.g., hair strands) and scales reliably to long clips. Impressive, code announced💙

👉Review https://lnkd.in/p/dj3_zE3P
👉Paper https://arxiv.org/pdf/2610.26857
👉Project https://cvlab-kaist.github.io/VideoMaMa-PlusPlus/
👉Repo https://github.com/cvlab-kaist/VideoMaMa-PlusPlus

🥭Agentic Manga->Animation🥭

👉Manga2Animate: turning existing manga pages into coherent animation that stays faithful to the original manga. Impressive results, best with sound. Code released.💙

👉Review https://lnkd.in/p/dHpdPAj3
👉Paper https://arxiv.org/pdf/2610.26851
👉Project https://annchenn.github.io/Manga2Animate/
👉Repo https://github.com/annchenn/Manga2Animate

🍵3D from Single Human RGB🍵

👉Real-to-semantic-to-real pipeline that turns one monocular RGB video of a human demonstration into thousands of robot demonstrations. Code TBA💙

👉Review https://lnkd.in/p/dt-B2QMt
👉Paper https://arxiv.org/pdf/2610.28880
👉Project https://real2sem2real.github.io/
👉Repo TBA

🧊 SOTA Foundation Pose 🧊

👉From one segmented RGB-D observation, directly predict 6D pose, metric size, and camera-frame completed shape. Repo under Apace 2.0💙

👉Review https://lnkd.in/p/dXpXt7Pz
👉Paper https://arxiv.org/pdf/2610.31623
👉Project https://jrryzh.github.io/F2G-Pose/
👉Repo https://github.com/jrryzh/F2G-Pose

🫀 SKIM V2 is out🫀

👉SOMA recovers muscle deformation over time from surface observations captured by multi-view RGB cameras, the first approach of this kind. SKIM v2: a new soft-tissue deformation dataset. Code under MIT, data & weights NC💙

👉Review https://lnkd.in/p/dB4XS7vB
👉Paper https://lnkd.in/d64S_qxj
👉Project https://lnkd.in/dWtavYGn
👉Repo https://lnkd.in/dXNJqY6N