Ismini Lourentzou

I am honored and grateful to receive an 𝗡𝗦𝗙 𝗖𝗔𝗥𝗘𝗘𝗥 award “𝘚𝘩𝘢𝘱𝘪𝘯𝘨 𝘌𝘮𝘣𝘰𝘥𝘪𝘦𝘥 𝘐𝘯𝘵𝘦𝘭𝘭𝘪𝘨𝘦𝘯𝘤𝘦 𝘛𝘩𝘳𝘰𝘶𝘨𝘩 𝘓𝘢𝘯𝘨𝘶𝘢𝘨𝘦-𝘎𝘶𝘪𝘥𝘦𝘥 𝘐𝘯𝘵𝘳𝘰𝘴𝘱𝘦𝘤𝘵𝘪𝘰𝘯.” 🎉 This project will support our work in the 𝗣𝗟𝗔𝗡 𝗟𝗮𝗯 toward more reliable, self-improving embodied AI systems. I am deeply thankful to @NSF, my students, whose creativity and hard work inspire me, as well as my collaborators, mentors, and @iSchoolUI colleagues for their support. I am also grateful to my former academic home @VT_CS where this work began to take shape. Excited for the work ahead! Learn more about our lab 👉 plan-lab.github.io #NSFCAREER #NSFfunded #EmbodiedAI #MultimodalAI #ResponsibleAI #iSchoolUI #PLANLab

@Jiaqi_Ma_ Thank you so much @Jiaqi_Ma_!

@Ismini_L Congratulations!!

I am honored and grateful to receive an 𝗡𝗦𝗙 𝗖𝗔𝗥𝗘𝗘𝗥 award “𝘚𝘩𝘢𝘱𝘪𝘯𝘨 𝘌𝘮𝘣𝘰𝘥𝘪𝘦𝘥 𝘐𝘯𝘵𝘦𝘭𝘭𝘪𝘨𝘦𝘯𝘤𝘦 𝘛𝘩𝘳𝘰𝘶𝘨𝘩 𝘓𝘢𝘯𝘨𝘶𝘢𝘨𝘦-𝘎𝘶𝘪𝘥𝘦𝘥 𝘐𝘯𝘵𝘳𝘰𝘴𝘱𝘦𝘤𝘵𝘪𝘰𝘯.” 🎉 This project will support our work in the 𝗣𝗟𝗔𝗡 𝗟𝗮𝗯 toward more reliable, self-improving embodied AI systems. I am deeply thankful to @NSF, my students, whose creativity and hard work inspire me, as well as my collaborators, mentors, and @iSchoolUI colleagues for their support. I am also grateful to my former academic home @VT_CS where this work began to take shape. Excited for the work ahead! Learn more about our lab 👉 plan-lab.github.io #NSFCAREER #NSFfunded #EmbodiedAI #MultimodalAI #ResponsibleAI #iSchoolUI #PLANLab

@yaoyaoliu1 Thank you @yaoyaoliu1 !

@Ismini_L Congrats! 🎉

@MagicaRobotics Thank you @MagicaRobotics!

@Ismini_L Congratulations!!! Well deserved!

🎉𝙿𝙻𝙰𝙽 𝙻𝚊𝚋 has two papers accepted to @icmlconf ! 🔥 𝐁𝐞𝐬𝐭 𝐨𝐟 𝐁𝐨𝐭𝐡 𝐖𝐨𝐫𝐥𝐝𝐬: 𝐌𝐮𝐥𝐭𝐢𝐦𝐨𝐝𝐚𝐥 𝐑𝐞𝐚𝐬𝐨𝐧𝐢𝐧𝐠 𝐚𝐧𝐝 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧 𝐯𝐢𝐚 𝐔𝐧𝐢𝐟𝐢𝐞𝐝 𝐃𝐢𝐬𝐜𝐫𝐞𝐭𝐞 𝐅𝐥𝐨𝐰 𝐌𝐚𝐭𝐜𝐡𝐢𝐧𝐠 Onkar Kishor Susladkar, Tushar Prakash, Gayatri Deshmukh, Kiet A. Nguyen, Jiaxun Zhang, Adheesh Sunil Juvekar, Tianshu Bao, Lin Chai, Sparsh Mittal, Inderjit S. Dhillon, Ismini Lourentzou 🔥 𝐋𝐚𝐓𝐭𝐄-𝐅𝐥𝐨𝐰: 𝐋𝐚𝐲𝐞𝐫𝐰𝐢𝐬𝐞 𝐓𝐢𝐦𝐞𝐬𝐭𝐞𝐩-𝐄𝐱𝐩𝐞𝐫𝐭 𝐅𝐥𝐨𝐰-𝐛𝐚𝐬𝐞𝐝 𝐓𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦𝐞𝐫𝐬 Ying Shen^, Zhiyang Xu^, Jiuhai Chen, Shizhe Diao, Jiaxin Zhang, Yuguang Yao, Joy Rimchala, Ismini Lourentzou*, Lifu Huang* Congratulations to @onkar_s_13, @YingShen_ys, all students, and collaborators! 👏🏿👏🏽👏🏻👏 #ICML2026 #GenerativeAI #DiffusionModels #MachineLearning

Check out our #CVPR2026 papers from PLAN Lab: 6 main conference papers + 1 Findings 🎉with more open-source releases coming soon! plan-lab.github.io/publications.h… Our #CVPR2026 work advances physically grounded, controllable multimodal intelligence 🔥physics-aware video generation, 🔥progress-aware VLAs for robotic manipulation, 🔥part-aware 3DGS for object articulation, 🔥new counterfactual benchmark for referring and reasoning segmentation hallucinations 🔥video tokenization, and 🔥reward-guided image generation

𝐏𝐲𝐫𝐚𝐓𝐨𝐤 generates high-resolution 4K/8K reconstructions while also remaining text-aware for grounded video understanding and generation tasks.

🚀 Excited to share our @CVPR 2026 paper 𝐏𝐲𝐫𝐚𝐓𝐨𝐤: Language-Aligned Pyramidal Tokenizer for Video Understanding and Generation! Work led by @onkar_s_13 👏 🔥 𝐏𝐲𝐫𝐚𝐓𝐨𝐤 delivers SOTA video reconstruction scaling to 4K/8K resolutions, and yields zero-shot performance on video segmentation, temporal action localization, and video understanding. Code and model weights open-sourced 👇 🌐 lnkd.in/gHwTe3GZ ⚙️ lnkd.in/giwb6GPw 🤗 lnkd.in/gCZ6gSyA #CVPR2026 #ComputerVision #MultimodalAI #ArtificialIntelligence

Can we mitigate embodied hallucinations and improve grounding in 3D-LLM agents without retraining?🤔 In our @CVPR paper, we propose 3D-VCD, a training-free, inference-time approach that uses contrastive decoding over 3D scene perturbations. 🤗 huggingface.co/papers/2604.08… 🌐 plan-lab.github.io/projects/3d-vcd #EmbodiedAI #3DLLMs #VisionLanguage #CVPR2026 #CVPR

PBench looks cool! Check out our HalluSegBench work (#CVPR2026 Findings) plan-lab.github.io/projects/hallu… We released a counterfactual segmentation reasoning benchmark in which models must segment the factual image and refuse to segment the counterfactual. HalluSegBench covers a wide range of counterfactual pairs + ✨New eval metrics included

We have negatives per prompt at each level, but did not release then as Saco covers this aspect well. Level 5, the task involves symbolic reasoning: given an area to be cleaned, the model must identify the dirt and reason about which parts of the floor require cleaning (for FP v2)

Vision Banana's first author replies to the criticism My summary and thoughts 🧵 SAM 3 came with SA-Co, a segmentation dataset that contains also negatives, e.g. given a photo of dogs, the text asks to segment cats (see colander below). The output should be an empty mask [1/N]

🎥✨ Most video generation models produce visually plausible content, but fail when it comes to actual physical dynamics 🎯Our #CVPR2026 Phantom work explicitly models latent physics alongside visual generation 👻Phantom learns: ⚡ visual trajectories (what we see) ⚡ physical dynamics (what governs motion) 🔀 through a dual-branch architecture with cross-attention Fantastic work led by @YingShen_ys 👏 🤗 huggingface.co/papers/2604.08… 🌐 plan-lab.github.io/projects/phant… #AI #VideoGeneration #GenerativeAI #WorldModels #ComputerVision #MultimodalAI #CVPR

Introducing Phantom 👻 — a Physics-Infused Video Generation model that jointly models visual content and latent physical dynamics without requiring explicit specification of complex physical properties. 👇🧵[1/3] 🌐 plan-lab.github.io/projects/phant… 📰 huggingface.co/papers/2604.08…

🚀 Excited to share our CVPR 2026 paper: RewardFlow: Generate Images by Optimizing What You Reward 🖼️ RewardFlow is an inversion-free framework for image editing and compositional generation that steers pretrained diffusion and flow-matching models at inference.

📢 We invite submissions to the @icmlconf 2026 Workshop From Frames to Stories (F2S): Toward Reliable, Controllable, and Trustworthy Long-Horizon Video Generation ✨The workshop brings together researchers advancing long-horizon video generation, with a focus on long-context modeling, state representation, interactive control, and trustworthy evaluation. 🎬 We welcome submissions of recent work in related areas, including both published and unpublished papers. 📅 Submission deadline: April 30, 2026 (AoE) 🌐 Website: icml2026-f2s-workshop.github.io #VideoGeneration #GenerativeAI #ICML2026 #ICML