Visual-Intelligence

Prof. Hao Su delivered a keynote speech at China3DV 2026 on April 17. The key to next-generation technologies is developing intervenable interactions, and AI systems are capable of successfully acting in unknown physical environments. #HaoSu #ai #machineintelligence

🎉🎉🎉Visual Intelligence is now in ESCI ! A huge thank you to our authors, reviewers, and readers for your support. We are excited to reach this major milestone! CiteScoreTracker 2025 10.6 All papers since 2023 will be indexed! link.springer.com/journal/44267/…

🔥The authors from Nanjing University of Science and Technology and Institute of Automation, CAS, introduce HiCache, a hierarchical timestep-aware caching framework for accelerating Diffusion Transformers. 🔗link.springer.com/article/10.100… #Generationacceleration @SpringerEng

🔥Authors from ShanghaiTech University proposed a comprehensive dynamic Gaussian processing framework that provides a complete, end-to-end pipeline. 🔗link.springer.com/article/10.100… #3DGaussianSplatting #VolumetricVideo @SpringerEng

🔥 Check out this T2M framework! It introduces a temporal consistency-aware spatial VQ-VAE for cross-sequence temporal alignment, coupled with a masked motion transformer for text-conditioned motion generation. link.springer.com/article/10.100… @SpringerEng #T2Mgeneration

This paper proposed segment anything model for drivable area segmentation termed as DAS-SAM, a novel efficient adaption framework that fine-tunes SAM towards DAS. link.springer.com/article/10.100… #SAM @SpringerEng

Join the 1st Workshop on Generative AI for Sign Language (GenSign) at CVPR 2026 @CVPR. 🚀 Paper submissions are NOW OPEN! We welcome papers on sign language processing, human-centric generative models, datasets/benchmarks, and ethics. 📌 Proceedings track DDL: Mar 14, 2026 (AoE) 📌 Non-proceedings track DDL: Apr 4, 2026 (AoE) 🌐 genai4sl.github.io ✉️ gensign.workshop@gmail.com #CVPR2026 #Sign_language #Human_centric #Benchmarks #Workshop

🔥Check out a new method of image dehazing. The authors proposed a synergic expert modulation (SEM) mechanism to explicitly model context information. 🔗link.springer.com/article/10.100… #dehaze @SpringerEng

HDVS online! The authors present a novel framework called heterogeneous dual-branch voting supervision (HDVS), which is designed to enhance the reliability of pseudo-labels and mitigate the issues arising from pseudo-labeling. link.springer.com/article/10.100… @SpringerEng

🔥Check out this paper! It proves that SAM2 can be a strong encoder for U-shaped segmentation models. The authors proposed a simple but effective framework, termed SAM2-UNet, for versatile image segmentation. link.springer.com/article/10.100… @SpringerEng #SAM2 #UNet

🔥Check out the NavGemini, a navigation system based entirely on the newly developed multi-modal LLM, Gemini-Pro-Vision. link.springer.com/article/10.100… #Visionandlanguagenavigation #LLM #EmbodiedAI @SpringerEng

🔥Check out the first visual-guided HOI (VG-HOI) detection task for detecting unseen HOI categories using a few guidance images. link.springer.com/article/10.100… #Humanobjectinteraction #Visualguidance @SpringerEng

✨Check out this paper. It systematically investigates the capability of deep generative models such as DragGAN, DragDiffusion, and DragonDiffusion for eye direction transformation. link.springer.com/article/10.100… #Eyedirectiontransformation @SpringerEng

@felix1987_ @JinaAI_ Check out this survey! The authors from Shanghai Jiao Tong University provide a comprehensive and systematic review of the current state of efficient multimodal large language models. link.springer.com/article/10.100…

✈️to ICLR'25 at Singapore. @JinaAI_ have two works (JinaCLIP-v2, and ReaderLM-v2) to present at the SCI-FM workshop. Looking forward to meeting you all and discussing efficient (small)LLMs, MLLM at #ICLR25

@shoubin621 Check out this survey! The authors from Shanghai Jiao Tong University provide a comprehensive and systematic review of the current state of efficient multimodal large language models. link.springer.com/article/10.100…

On compositional video reasoning, CREMA is significantly efficient (with ~97% fewer trainable parameters) yet outperforms strong MLLM baselines on both 3D-associated (table1) and audio-associated (table2) video reasoning.

How to efficiently & flexibly inject diverse new modalities (RGB+audio/3D/depth/flow/...) to improve Video Reasoning? Introducing ☕️ CREMA, a substantially efficient (~3% trainable params) & modular fusion framework (w/o complex architecture changes) arxiv.org/abs/2402.05889 🧵

🔥Now online. The authors from Shanghai Jiao Tong University provide a comprehensive and systematic review of the current state of efficient multimodal large language models. link.springer.com/article/10.100… @SpringerEng #efficientMLLM

🔥Check out the environment dynamics decomposition (ED2), a novel world model construction framework that models the environment in a decomposing manner. link.springer.com/article/10.100… #Reinforcementlearning @SpringerEng

🔥 Check out this paper. It introduced a novel concept of deepfake detection calibration, which focuses on aligning predicted probability with actual distribution to alleviate overconfidence and estimate the uncertainty. link.springer.com/article/10.100… #Deepfakedetection @SpringerEng

Check out GazeLLM, the first zero-shot large language model (LLM) boosted framework for gaze target reasoning. Authors from Beihang University. link.springer.com/article/10.100… #Gazetargetdetection #socialgazeprediction #LLM @SpringerEng

Cover Story: The research team proposed DriveMLM, an LLM-based autonomous driving framework that can perform close-loop autonomous driving in realistic simulators. Visual Intelligence, 2025, Volume 3, Article no. 22. Learn more here: link.springer.com/article/10.100… @SpringerEng