🔌 BodyMAP: human body & pressure 🔌

👉#Nvidia (+CMU) unveils BodyMAP, the new SOTA in predicting body mesh (3D pose & shape) and 3D applied pressure on the human body. Source Code released, Dataset coming 💙

👉Review https://t.ly/8926S
👉Project bodymap3d.github.io/
👉Paper https://lnkd.in/gCxH4ev3
👉Code https://lnkd.in/gaifdy3q

📈Gradient Boosting Reinforcement Learning📈

👉#Nvidia unveils GBRL, a framework that extends the advantages of Gradient Boosting Trees to the RL domain. GBRL adapts the power of Gradient Boosting Trees to the unique challenges of RL environments, including non-stationarity and absence of predefined targets. Code released💙

👉Review https://t.ly/zv9pl
👉Paper https://arxiv.org/pdf/2407.08250
👉Code https://github.com/NVlabs/gbrl

🛳️ EVER Ellipsoid Rendering 🛳️

👉UCSD & Google present EVER, a novel method for real-time differentiable emission-only volume rendering. Unlike 3DGS it does not suffer from popping artifacts and view dependent density, achieving ∼30 FPS at 720p on #NVIDIA RTX4090.

👉Review https://t.ly/zAfGU
👉Paper arxiv.org/pdf/2410.01804
👉Project half-potato.gitlab.io/posts/ever/

🪞Robo-Emulation via Video Imitation🪞

👉OKAMI (UT & #Nvidia) is a novel foundation method that generates a manipulation plan from a single RGB-D video and derives a policy for execution.

👉Review https://t.ly/_N29-
👉Paper arxiv.org/pdf/2410.11792
👉Project https://lnkd.in/d6bHF_-s

🧞‍♂️Omni-RGPT: SOTA MLLM Understanding🧞‍♂️

👉 #NVIDIA presents Omni-RGPT, MLLM for region-level comprehension for both images & videos. New SOTA on image/video-based commonsense reasoning.

👉Review https://t.ly/KHnQ7
👉Paper arxiv.org/pdf/2501.08326
👉Project miranheo.github.io/omni-rgpt/
👉Repo TBA soon

🌈 #Nvidia Foundation ZS-Stereo 🌈

👉Nvidia unveils FoundationStereo, a foundation model for stereo depth estimation with strong zero-shot generalization. In addition, a large-scale (1M stereo pairs) synthetic training dataset featuring large diversity and high photorealism. Code, model & dataset to be released💙

👉Review https://t.ly/rfBr5
👉Paper arxiv.org/pdf/2501.09898
👉Project nvlabs.github.io/FoundationStereo/
👉Repo github.com/NVlabs/FoundationStereo/tree/master

🥛HAMSTER: Hierarchical VLA Manipulation🥛

👉#Nvidia unveils HAMSTER: novel Hierarchical VLA architecture to enable robotic manipulation with semantic, visual & geometric generalization trained on easy to collect, off-domain data. Source Code announced💙

👉Review https://t.ly/2yXaY
👉Paper https://arxiv.org/pdf/2502.05485
👉Project https://hamster-robot.github.io/
👉Repo TBA

🌈Unified Low-Level 4D Vision🌈

👉#Nvidia L4P is a novel feedforward, general-purpose, architecture to solve low-level 4D perception tasks in a unified framework. L4P combines a ViTbased backbone with per-task heads that are lightweight and therefore do not require extensive training. One backbone - many SOTAs. Code announced 💙

👉Review https://t.ly/04DGj
👉Paper arxiv.org/pdf/2502.13078
👉Project research.nvidia.com/labs/lpr/l4p/
👉Repo TBA

👽Neural-Free Sparse Voxels Rasterization👽

👉#Nvidia unveils a novel efficient radiance field rendering algorithm that incorporates a rasterization process on adaptive sparse voxels without neural networks or 3D Gaussians. Code released (custom license)💙

👉Review https://t.ly/Nh_ic
👉Paper https://lnkd.in/g8k8Zs6R
👉Project https://lnkd.in/gR-bD4Wx
👉Repo https://lnkd.in/gNHX-w4t

🙀3D MultiModal Memory🙀

👉M3 is a novel framework by UCSD & #NVIDIA for rendering 3D scenes w/ RGB & foundation model embeddings. Rich spatial & semantic understanding via novel memory system designed to retain multimodal info through videos

👉Review https://t.ly/OrXZO
👉Paper arxiv.org/pdf/2503.16413
👉Project https://lnkd.in/dXAZ97KH
👉Repo https://lnkd.in/dWvunCET

🦎 Scaling Vision to 4K🦎

👉PS3 by #Nvidia (+UC Berkeley) to scale-up CLIP-style vision pre-training to 4K with *near-constant* cost. Encoding LR global image and selectively processes only informative HR regions. Impressive work. Code/weights & 🤗 announced💙

👉Review https://t.ly/WN479
👉Paper https://lnkd.in/ddWq8UpX
👉Project https://lnkd.in/dMkTY8-k
👉Repo https://lnkd.in/d9YSB6yv

🍏PartField #3D Part Segmentation🍏

👉#Nvidia unveils PartField, a FFW approach for learning part-based 3D features, which captures the general concept of parts and their hierarchy. Suitable for single-shape decomposition, co-segm., correspondence & more. Code & Models released under Nvidia License💙

👉Review https://t.ly/fGb2O
👉Paper https://lnkd.in/dGeyKSzG
👉Code https://lnkd.in/dbe57XGH
👉Project https://lnkd.in/dhEgf7X2

🦧 #Nvidia Describe Anything 🦧

👉Nvidia unveils Describe Anything Model (DAM) the new SOTA in generating detailed descriptions for user-specified regions in images/videos, marked by points, boxes, scribbles, or masks. Repo under Apache, Dataset available and live demo on 🤗

👉Review https://t.ly/la4JD
👉Paper https://lnkd.in/dZh82xtV
👉Project https://lnkd.in/dcv9V2ZF
👉Repo https://lnkd.in/dJB9Ehtb
🤗Demo https://lnkd.in/dXDb2MWU

🍏#Nvidia Dynamic Pose 🍏

👉Nvidia unveils DynPose-100K, the largest dataset of dynamic Internet videos annotated with camera poses. Dataset released under Nvidia license💙

👉Review https://t.ly/wrcb0
👉Paper https://lnkd.in/dycGjAyy
👉Project https://lnkd.in/dDZ2Ej_Q
🤗Data https://lnkd.in/d8yUSB7m

🧞‍♀️GENMO: Generalist Human Motion 🧞‍♀️

👉#Nvidia presents GENMO, a unified Generalist Model for Human Motion that bridges motion estimation and generation in a single framework. Conditioning on videos, 2D keypoints, text, music, and 3D keyframes. No code at the moment🥲

👉Review https://t.ly/Q5T_Y
👉Paper https://lnkd.in/ds36BY49
👉Project https://lnkd.in/dAYHhuFU

🧤Diffusive Hand from Signs🧤

👉LIGM + #NVIDIA unveil a novel generative model of 3D hand motions from Sign Language Data. Motion characteristics such as handshapes, locations, finger, hand & arm movements. Code, Models & Data to be released 💙

👉Review https://t.ly/HonX_
👉Paper https://arxiv.org/pdf/2508.15902
👉Project https://imagine.enpc.fr/~leore.bensabath/HandMDM/
👉Data drive.google.com/drive/u/1/folders/1BLsu2hAqhAJ_gnGb9TNXW7MLiSuSEzEj
👉Repo TBA

🛡️3D Prompted Vision-LLM🛡️

👉#Nvidia unveils SR-3D, a novel aware vision-language model that connects single-view 2D images and multi-view 3D data through a shared visual token space. Flexible region prompting, allowing users to annotate regions with bounding boxes, segmentation masks on any frame, or directly in 3D, without the need for exhaustive multi-frame labeling. Code & Dataset announced💙

👉Review https://t.ly/5Y2c5
👉Paper https://arxiv.org/pdf/2509.13317
👉Project https://www.anjiecheng.me/sr3d
👉Repo TBA

A few “leaks” for you from the #Nvidia presentation I’m right now in Milan. Impressive stuff.

Ps: sorry for the shitty quality of the pics ♥️

🤖Real-time Interactive Video🤖

👉LONGLIVE by #Nvidia is a frame-level autoregressive framework for real-time & interactive long video generation. LONGLIVE accepts sequential user prompts and generates corresponding videos in real time. Repo under non-commercial license💙

👉Review https://t.ly/jJkdY
👉Paper arxiv.org/pdf/2509.22622
👉Project nvlabs.github.io/LongLive/
👉Repo github.com/NVlabs/LongLive
🤗huggingface.co/Efficient-Large-Model/LongLive-1.3B