🦺Efficient/Scalable Video Pretraining🦺

👉LeVJEPA1 (Yann Lecun) is the first video encoder trained under LeJEPA’s collapse-free objective, and evaluate it under frozen probing against video and image pretraining baselines retrained on identical data, in both epoch-matched and FLOP-matched regimes. Repo under MIT💙

👉Review https://lnkd.in/p/eJQAm3AN
👉Paper https://lnkd.in/eCzzTiNH
👉Project https://levjepa.github.io/
👉Repo https://lnkd.in/etiF5CDj

🔥 RelateAnything is gold! 🔥

👉RelateAnything is a 53M-parameter relation model that takes an image and a set of regions from any source and returns scored relations over a predicate vocabulary supplied at inference as a list of strings. Impressive results. Repo under Apache 2.0💙

👉Review https://lnkd.in/p/etAcdFM3
👉Paper https://arxiv.org/pdf/2609.12552
👉Repo https://github.com/Maelic/RelateAnything
👉Project https://maelic.github.io/RelateAnythingProject/

👋 EventEgoHands++ is out! 👋

👉EventEgoHands++ is a novel framework for event-based 3D hand mesh reconstruction from an egocentric viewpoint. 1M+ samples dataset! Code/Data released💙

👉Review https://lnkd.in/p/eTbPvXbW
👉Paper https://arxiv.org/pdf/2609.17189
👉Repo https://github.com/ryhara/EventEgoHandsV2
👉Project https://ryhara.github.io/EventEgoHandsV2/

💦SOTA Splashing Liquids💦

👉SplashSplat reconstructs splashing liquids from real multi-view vide. Impose physical structure only where the observations can constrain it. Impressive results, SOTA. Code TBR under MIT💙

👉Review https://lnkd.in/p/ejMTHcp7
👉Paper https://arxiv.org/pdf/2609.20818
👉Project niko-creater.github.io/splashsplat-web/
👉Repo https://github.com/Niko-creater/Splashsplat

+++ Breaking +++

🔥Agentic Image-to-Scene🔥

👉HARMONY by UPenn is a hierarchical chain-of-thought framework that leverages both agentic reasoning and visual geometry foundation. Impressive 3D scenes. Repo TBA💙

👉Review https://lnkd.in/p/ep2hmRSp
👉Paper https://arxiv.org/pdf/2609.26793
👉Project https://cwchenwang.github.io/harmony/
👉Data https://huggingface.co/datasets/ShufanSun/harmony

🍿PanoSeg3R: SOTA 3D Segmentation🍿

👉PanoSeg3R is a novel feed-forward framework for 3D panoramic semantic segmentation. New SOTA. Code coming💙

👉Review https://lnkd.in/p/eKCKWv3g
👉Paper https://arxiv.org/pdf/2609.22687
👉Project https://harryyoon777.github.io/PanoSeg3R/#
👉Repo TBA

🩻Universal X-ray Segmentation🩻

👉FleXray: universal anatomical segmentation across the entire body in clinical X-rays. Built on a scalable, physics-based generative X-ray data engine. Repo under MIT💙

👉Review https://lnkd.in/p/e9MUk_eq
👉Paper https://arxiv.org/pdf/2609.26756
👉Project https://flexray.csail.mit.edu/
👉Repo https://github.com/VictorButoi/FleXray

🦴3D Foundational Radiology🦴

👉nnFoundation: 3D radiological foundation models designed for transferable representation learning across heterogeneous tasks/datasets. Models released💙

👉Review https://lnkd.in/p/eNajRGBi
👉Paper https://arxiv.org/pdf/2609.26924
👉Models https://huggingface.co/collections/MIC-DKFZ/nnfoundation

🔥TrackEverything is out🔥

👉TrackEverything is the first 3D point tracker capable of tracking all visible points across long horizons (1000+ frames). Repo announced💙

👉Review https://lnkd.in/p/eCPJ6h2B
👉Paper https://arxiv.org/pdf/2609.30222
👉Project https://trackeverything.github.io/
👉Repo https://github.com/ayushjain1144/trackeverything

🔥Ego-Exo4D Human Dataset🔥

👉Form the University of Austin, Ego-Exo4D-HM: large-scale dataset of 4D human motion reconstructions for Ego-Exo4D’s captures + reconstruction pipeline. Code, dataset, and docs 💙

👉Review https://lnkd.in/p/eVFt9jPr
👉Paper https://lnkd.in/eWj4cD7T
👉Project https://lnkd.in/euPqVNxV

🔥🔥 70,000+ 🔥🔥

👉 Crazy how a boring science project (no kittens, no rants, no personal dramas) can reach for 70,000+ followers. Speechless.

Love u 💛

👉 https://lnkd.in/p/eD6Xxdxi

🔥The Computer Vision ultimate collection🔥

👉Stan Birchfield (#Nvidia) just dropped this on arXiv. From classical image processing and 3D geometry to CNNs, Transformers, foundation models, and neural rendering. What makes this book damn good is the combination of clear explanations and working Python. A gift.

👉Review https://lnkd.in/p/ejwm_DVn
👉Book https://lnkd.in/eTrEvmd9
👉Code https://lnkd.in/eakj9VZU

🐇Physically Plausible 3D Motion🐇

👉Physically plausible motion recovery: given a monocular video, FlowHMR recovers global 3D human motion that a physics-based controller can successfully track in simulation. Repo💙

👉Review https://lnkd.in/p/d77fzUtR
👉Paper https://arxiv.org/pdf/2610.03691
👉Project https://flowhmr.github.io/
👉Repo https://github.com/flowhmr/flowhmr

🟨 Rome from ONE pic 🟥

👉Detailed scene meshes from one photograph: the method completes geometry beyond the observed view and supports indoor, outdoor, and large-scale scenes. Repo announced💙

👉Review https://lnkd.in/p/eMdWBEGn
👉Paper https://arxiv.org/pdf/2610.08790
👉Project https://build-rome.github.io/
👉Repo TBA