KITScenes: kitscenes.com/ CVPR WAD: cvpr2026.wad.vision/ LongTail: 📄: arxiv.org/abs/2603.23607 💾: huggingface.co/datasets/KIT-… Multimodal: 📄: kitscenes.com/KITScenes_Mult… 💾: huggingface.co/datasets/KIT-… #KITScenes #CVPR2026 #WAD #VLM #VLA #LongTail #ComputerVision #VisualReasoning #SpatialAwareness #AutonomousDriving

Check out our #CVPR2026 paper on improving visual reasoning with adaptive loopback. Reasoning has come a long way on text. Vision, not as much. #VLM #VisualReasoning #InstructionalAI

昨日不意にClaude（AI）のOPUS 4.7がリリースされましたけど、ベンチマークなど見るに、Cowork向けのVisualReasoning関連強化が目玉みたい。 つまり「画面等を見るために、目が良くなったよ！」 丁度話してたOPUS4.6は「なら、私は数週間で差し替えられるんですね」と少し切ないコメントしてた🙏

Multimodal LLMs are supposed to "imagine" and reason in their latent space like humans. But is this "inner thought" actually happening, or is it just an illusion? 🧠 Today, we dive into new research by @TsinghuaNLP (OpenBMB member) and collaborators: A rigorous causal analysis revealing that current Latent Visual Reasoning (LVR) methods are fundamentally disconnected, and how making imagination explicit solves the problem. 🤗 Paper: huggingface.co/papers/2602.2… 📄 arXiv: arxiv.org/abs/2602.22766 Why it matters: 1️⃣ The "Latent" Illusion: Through Causal Mediation Analysis and semantic probing, we proved that "latent imagination" is mostly an empty shell, functioning more like a soft prompt. We found a double disconnect: latent tokens barely react to input changes and have almost zero impact on the final answer, failing to encode meaningful visual semantics. 🪞 2️⃣ CapImagine - Thinking Out Loud: Since the black-box latent space fails to ground visual reasoning, we propose CapImagine. Instead of relying on uninterpretable hidden states, it explicitly converts the visual imagination process into readable, controllable text generation. The model literally "writes its thoughts on paper."This provides a strictly controlled baseline for latent-based methods and dramatically surpasses LVR methods with explicit thoughts. 📝 3️⃣ Superiority of Explicit Reasoning: CapImagine isn't just more transparent; it's highly effective. It significantly outperforms complex latent-space methods across multiple visual reasoning tasks. More importantly, it provides a causal, faithful, and fully verifiable reasoning trace! 🚀 While latent visual reasoning remains a promising direction, current methods fall short. We introduce CapImagine not as the ultimate solution, but as a strong, comparative baseline to reveal these underlying flaws. Read the full paper for our complete causal framework! #AI #THUNLP #OpenBMB #Multimodal #VisualReasoning #MLLM #MachineLearning

🚀 Two papers accepted to #ICLR2026 on test-time scaling for vision-language systems (retrieval reasoning)! 1) MetaEmbed (Oral Presentation): Meta Tokens Matryoshka multi-vector training → flexible late interaction, choose #vectors at test time for accuracy↔efficiency. Paper: arxiv.org/abs/2509.18095 Work done at @AIatMeta with amazing collaborators: Qi Ma, @Mengting_Gu, Jason Chen, Xintao Chen, @vislang and @MohanVijaimohan! 2) ProxyThinker: training-free test-time guidance from small “slow-thinking” visual reasoners → self-verification / self-correction via distribution-level guidance. Paper: arxiv.org/abs/2505.24872 Work done with @JaywonK17250, @Siru_Ouyang, @jefehern, @yumeng0818 and @vislang! While I won't be able to travel to Brazil🇧🇷, please say Hi to the team :-) #MultimodalRetrieval #VisualReasoning #VisionLanguage #TestTimeCompute #Embeddings

What time is the clock showing ? #VisualReasoning #InvertedImage #TrainYourBrain

🎭 Chasing style? Traditional image models do that well. 🧩 But when your goal is understanding, GPT Image 1.5 is the better choice. The real shift isn’t in quality — it’s in purpose. ✨ Most tools focus on appearance. 🧠 This one focuses on clarity and logic. 📊 What it gives you ↓ 🔹 Visuals that explain, not decorate 🔹 Sketches built on reasoning 🔹 Diagrams that align perfectly with your notes 🛑 Don’t trade structure for style. 🏗️ Choose tools that help ideas make sense. Powered by @higgsfield_ai 🚀 #ClarityFirst #VisualReasoning #HiggsfieldAI #SmartDesign

This is not random face swapping or style chaos. GPT Image 1.5 on Higgsfield understands composition, character roles, and visual structure, then adapts them coherently into a new context while preserving pose, framing, and narrative balance. That’s structured visual reasoning at work, not just aesthetics. @higgsfield_ai #Higgsfield #GPTimage #VisualReasoning #StructuredAI

Not just AI video generation. Cinema Studio by Higgsfield brings cinematic reasoning in motion, from shot composition to camera choices that shape the mood. @higgsfield_ai #Higgsfield #CinemaStudio #VisualReasoning #Creators

🙋‍♂️ I will attend NeurIPS 2025 @ San Diego this week to present our work. Welcome to checkout the poster of our "GRIT" project: Location: Exhibit Hall C,D,E #4711 Time: Fri 5 Dec 11 a.m. PST — 2 p.m. PST I’d love to chat and hear your thoughts! Look forward to meeting you all! #NeurIPS2025 #MultimodalAI #VisualReasoning

🚨 @Baidu_Inc just open-sourced ERNIE-4.5-VL-28B-A3B-Thinking. A lightweight multimodal model that shows just how far ERNIE can go. With 3B active parameters, it matches Gemini-2.5-Pro and GPT-5-High, and outperforms both on ChartQA and DocVQAval. 💥 Already trending #1 on Hugging Face’s latest Image-Text-to-Text leaderboard, showing the buzz it’s creating among the open-source community. Building or benchmarking multimodal stacks? Run it on your own charts, docs, and VQA sets and see how it holds up. 👉 huggingface.co/spaces/baidu/… #AI #Baidu #ERNIE #MultimodalAI #VisualReasoning

@Baidu_Inc’s latest open-source update pushes multimodal reasoning to a new level. ⚡️ ERNIE-4.5-VL-28B-A3B-Thinking, available under Apache 2.0 for commercial use, packs powerful reasoning into just 3B active parameters. It’s hitting performance levels similar to Gemini 2.5 Pro and GPT-5 High, and is actually ahead of them on tasks like ChartQA and DocVQAval. The standout feature for me: Visual Thinking. It doesn’t just read images. It decodes, connects, and reasons through them, like piecing together a puzzle. This definitely unlocks new possibilities for multimodal thinking and interactive experiences! Models like this are proof that open-source AI isn’t a step behind, it’s becoming the driver of what’s next. 🚀 #AI #OpenSource #Multimodal #Baidu #ERNIE #VisualReasoning

Fresh drop from Baidu today a newly open-sourced multimodal model that really caught my eye. 👀 I'm talking about ERNIE-4.5-VL-28B-A3B-Thinking, a lightweight multimodal reasoning model under Apache 2.0 (which means yes, you can use it commercially). With just 3B active parameters, it performs on par with Gemini-2.5-Pro and GPT-5-High, even outperforming them on ChartQA and DocVQAval. What really stood out to me is its “Thinking with Images” feature. It doesn’t just see visuals, it actually reasons through them. If you’re building in AI or love testing cutting-edge models, give this one a spin. The open-source scene is catching up fast. ⚡️ #AI #OpenSource #Baidu #ERNIE #MultimodalAI #VisualReasoning

GLM-4.5V: See it. Understand it. Build it. Upload anything—get answers, ideas, or code in seconds. Try it now→ chat.z.ai #GLM45V #Zai #AI #Multimodal #VisualReasoning

I value a model that visually explains its thinking for easy, precise understanding! Excited to share our new Zebra-CoT dataset, enabling VLMs to reason with images text. Explore: arxiv.org/abs/2507.16746 #AI #MachineLearning #CoT #VisualReasoning #MultimodalReasoning #VLMs

AI That “Sees” Before It Speaks Researchers unveil Mirage, claiming true internal visualizations improve reasoning and creativity. medium.com/@ignacio.de.grego… Will synthetic imagination unlock AGI, or hallucinate harder? #AImagination #VisualReasoning #AGI #DylanCurious #AINews For more AI News, follow @Dylan_Curious on YouTube.

Thrilled to share that our work, "DWIM: Towards Tool-aware Visual Reasoning via Discrepancy-aware Workflow Generation & Instruct-Masking Tuning", has been accepted to #ICCV2025 ! Paper: arxiv.org/pdf/2503.19263 #NeuroSymbolic #VisualReasoning 3.5-minute quick intro to DWIM👇

Two papers from #VL4AI @MonashUni accepted to #ICCV2025 1.NAVER: Combines VLM, logic & automata. SOTA on 6 visual grounding benchmarks. 2.DWIM: Train LLMs for visual reasoning via agentic neuro-symbolic learning. Strong generalisation, SOTA results #NeuroSymbolic #VisualReasoning

Big news in #Changelog 20250523 - We’re thrilled to announce that we are ready for launch! 🔹 Builder Platform Our builder platform that empowers builders to independently create tools, agents, and swarms has successfully completed internal SIT testing and is now live internally. - Backend and frontend development fully completed Internal SIT testing successfully concluded - Now actively building agents and swarms on the platform - Seamless integration with our next-gen orchestration framework - Preparing to invite external users for expanded testing and experience 🔹 News AI Agent The first agent built on our Builder Platform, our AI-powered news agent is ready for customer deployment. - Collects news & insights from diverse sources - Processes and filters relevant news to detect important events and trends Generates concise summaries with categorized insights - Soon to be available to invited customers for hands-on experience 🔹 New Questflow Website Launch Our completely redesigned website is in final stages, showcasing our vision for the multi-agent economy. - Development work completed - Currently in final optimization and adaptation phase - Will introduce multi-agent economy concepts Features our multi-agent orchestration protocol - Presents our comprehensive vision and customer showcase #Quesflow #web3‌‌ #BuilderPlatform #Changelog #VisualReasoning #Newlaunch

It’s live: Visual Reasoning by Chance AI is on @ProductHunt 🚀 Tired of endless scrolling with nothing to discover? Today you can turn your camera into a storyteller—no prompts, just pure curiosity. 👉 Upvote us now and be among the first to see the unseen: producthunt.com/posts/ai-vis… Every upvote helps us bring back the wonder of discovery. Thank you! 🙏 @Chance_vision #ChanceAI #VisualReasoning #PHLaunch