After 1yr of Building VLMEvalKit now reaches 100+ Contributors On the journey of exploring LMM capabilities, we will go further

🚀 Introducing Aria-UI – a cutting-edge grounding LMM for GUI agents with a lightning-fast 3.9B parameters activated backbone! 🌐 Try it yourself: 📄 Project page: 📂 Explore on GitHub:

𝙊𝙪𝙧 𝙣𝙚𝙬𝙚𝙨𝙩, 𝙢𝙤𝙨𝙩 𝙘𝙤𝙢𝙥𝙧𝙚𝙝𝙚𝙣𝙨𝙞𝙫𝙚 𝙨𝙪𝙧𝙫𝙚𝙮 𝙤𝙣 𝙑𝙞𝙙𝙚𝙤 𝙌𝙪𝙖𝙡𝙞𝙩𝙮 𝘼𝙨𝙨𝙚𝙨𝙨𝙢𝙚𝙣𝙩—led by my legendary advisor, Alan Bovik, who has pioneered this field for over three decades, and myself, dedicated my (almost) entire PhD journey to this topic—𝒊𝒔 𝒏𝒐𝒘 𝒍𝒊𝒗𝒆 𝒐𝒏 𝒂𝒓𝑿𝒊𝒗! 📎 Paper: 🖥️ GitHub: In this work, we’ve curated a panoramic, deeply-researched view of the Video Quality Assessment (VQA) landscape. We cover the evolution from classic methods to cutting-edge deep learning solutions—offering a clear guide for both newcomers and seasoned experts. 𝐊𝐞𝐲 𝐡𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬 𝐢𝐧𝐜𝐥𝐮𝐝𝐞: 📚 A holistic categorization and analysis of existing VQA models, with insights into how techniques have evolved and where they’re headed. 🧐 A thorough look at subjective evaluation fundamentals, including major datasets and what they mean for real-world applications. 🤖 A deep dive into loss functions and architectural innovations, illuminating how modern frameworks are pushing the frontier of #VQA. 📊 Broad comparisons across emergent data types, shedding light on the importance of modeling spatiotemporal details and leveraging prior knowledge. 🎯 Real-world applications and future directions that underscore how these advancements can revolutionize streaming platforms, social media, and beyond. We hope this survey catalyzes new research avenues, encourages innovative solutions, and serves as a catalyst for the potential industry-university cooperation to foster fast and practical integration of such essential technologies into the social media, video streaming, or even the generative imagery/videography industry! 🚀 Dive in, share your thoughts, and let’s drive the future of #VQA together!

Gemini 2.0 Flash is out and it has lots of exciting things - Audio (multilingual)+image generation - Image editing - Multimodal real-time API - 2D/3D spatial understanding 🤯 - Great code capabilities Try it: Docs:

I'll be at #NeurIPS 2024. Join me at the LongVideoBench Poster (East Exhibit Hall A-C, #4611) on Wed, Dec 11, 4:30–7:30 PM. Let’s chat about all things related to multimodal research!