Happy to share our #SIGGRAPHAsia2024 technical paper - 𝗥𝗲𝗩𝗲𝗿𝘀𝗶𝗼𝗻 (3 Dec, Tue PM). It personalizes/captures the 𝘳𝘦𝘭𝘢𝘵𝘪𝘰𝘯 of entities from several exemplar images. While the well-known research (e.g., DreamBooth, TI ...) study 𝘢𝘱𝘱𝘦𝘢𝘳𝘢𝘯𝘤𝘦 𝘪𝘯𝘷𝘦𝘳𝘴𝘪𝘰𝘯, ReVersion took a step to explore 𝘳𝘦𝘭𝘢𝘵𝘪𝘰𝘯 𝘪𝘯𝘷𝘦𝘳𝘴𝘪𝘰𝘯 - an important yet under-explored pillar of the visual world.

RT @liuziwei7: 📢Representation Matters for Video Generation📢 #RepVideo is a representation-centric framework for T2V with cross-layer feat…

RT @_akhaliq: RepVideo Rethinking Cross-Layer Representation for Video Generation

RT @xunhuang1995: 🥇CausVid is ranked 1st on VBench (among officially evaluated video generation models). 🚀It is also several hundred times…

RT @ai_hakase_: 【VBench Video Arena が凄い！🎥】 ✎. FYIG: AI によって生成された動画を、なんと18万本以上も視聴できる「VBench Video Arena」が公開されました！👀 4…

RT @yen_chen_lin: For content creation, VBench (Huang et al., 2024) is a popular benchmark suite that evaluates video generation models on…

View the AI-generated videos here - No need to run code, set up environments, or download massive datasets. Thanks @_akhaliq for Gradio support, and kudos to @yinanhenevernan and @Zhang_Fan_ for the tremendous efforts in bringing this together! #VBench

RT @liuziwei7: 🚀AI Agent for GenAI Evaluation🚀 🤖Evaluation Agent🤖 is a multi-agent framework for efficient, dynamic, multi-round evaluatio…

RT @shulin_tian: Tired of waiting forever to evaluate your Gen models? 🚀 Meet 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 𝗔𝗴𝗲𝗻𝘁 🤖📊 – a fast, efficient, and promptable frame…

The 𝗩𝗕𝗲𝗻𝗰𝗵 𝗟𝗲𝗮𝗱𝗲𝗿𝗯𝗼𝗮𝗿𝗱 now features 50+ 𝘁𝗲𝘅𝘁-𝘁𝗼-𝘃𝗶𝗱𝗲𝗼 models, with 17 new additions since October, including 𝘏𝘶𝘯𝘺𝘶𝘢𝘯𝘝𝘪𝘥𝘦𝘰, 𝘓𝘛𝘟-𝘝𝘪𝘥𝘦𝘰, 𝘊𝘰𝘨𝘝𝘪𝘥𝘦𝘰𝘟-1.5𝘉, 𝘑𝘪𝘮𝘦𝘯𝘨, 𝘝𝘪𝘥𝘶, 𝘔𝘰𝘤𝘩𝘪-1, 𝘖𝘱𝘦𝘯𝘚𝘰𝘳𝘢𝘗𝘭𝘢𝘯-𝘝1.2 & 𝘝1.3, 𝘢𝘯𝘥 𝘔𝘪𝘯𝘪𝘔𝘢𝘹-𝘝𝘪𝘥𝘦𝘰-01, among others. For closed-source or API-based models, we welcome collaborative evaluations via the provided API and can conduct evaluations based on available resources. If you’re submitting final evaluation results directly, please include a project page link or relevant details as outlined in the submission portal. We strongly encourage submitters to share sampled videos used for VBench evaluation. These videos are stored in a publicly accessible space to ensure transparency and reproducibility. To maintain leaderboard quality, we periodically remove entries with unknown sources. If your model is missing or needs updates, don’t hesitate to reach out! #VBench #AIGC #T2V

Evaluating visual generative models is slow—sampling massive amounts of images/videos takes time, especially for diffusion models. Inspired by how humans "test and see" a few examples for quick judgment, we introduce 📊 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 𝗔𝗴𝗲𝗻𝘁 🤖: • Fast and efficient • Supports open-ended evaluation requests • Provides explanations for evaluation scores • Scales easily with additional metrics and tools

RT @Zhang_Fan_: 🎉 Excited to share Evaluation Agent! 🎉 An efficient, promptable evaluation framework for visual generative models, mimicki…

RT @qhnmoon: Unleash the resolution of your SDXL without cost. 🚀FreeScale🚀, a tuning-free method for higher-resolution visual generation, u…

RT @liuziwei7: 📢📢 #SIGGRAPHAsia2024 Welcome to check out our Multimodal GenAI work @SIGGRAPHAsia ** Reversion: Relation Inversion from Im…

📊 VBench++: Comprehensive and Versatile Benchmark Suite for Video Generative Models 🎞️ 🏛️ What VBench++ Evaluates: • 𝗧𝗲𝘅𝘁-𝘁𝗼-𝗩𝗶𝗱𝗲𝗼 (16 comprehensive dimensions) • 𝗜𝗺𝗮𝗴𝗲-𝘁𝗼-𝗩𝗶𝗱𝗲𝗼 (check out our size-adaptive Image Suite) • 𝗧𝗿𝘂𝘀𝘁𝘄𝗼𝗿𝘁𝗵𝗶𝗻𝗲𝘀𝘀 (fairness, bias, safety) 👁️ Extensive Human-Alignment Validation in every Dimension

RT @Hailuo_AI: 🏆 VBench Update 🏆 Exciting news! We’re thrilled to announce that we've secured the #1 ranking on both the #VBench and #VBenc…

RT @_tianxing: Interested in improving the inference quality of diffusion models? Checkout our #FreeInit poster at No.226 from 10:30am to…