生数科技联合清华推出sora竞品视频生成大模型Vidu,一键生成16秒高清大片
Vidu主要特点
时长16秒1080p视频
模拟真实世界
丰富的想象力
多镜头
时空一致性
目前来看只有时长与sora有差距,sora是一分钟,其他指标全部达到世界顶尖水平,从生数科技官网,我们可以清晰了解到,vidu团队掌握基础理论到产品的端到端的视频生成世界顶尖技术,一句话:
Vidu掌握核心科技
通用架构
·All are Worth Words: A ViT Backbone for Diffusion Models(CVPR 2023)
·UniDiffuser: One Transformer Fits All Distributions in Multi-Modal Diffusion at Scale (ICML 2023)
高速采样
·Analytic-DPM: an Analytic Estimate of the Optimal Reverse Variance in Diffusion Probabilistic Models (ICLR 2022)
·DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps (NeurIPS 2022)
·Estimating the Optimal Covariance with Imperfect Mean in Diffusion Probabilistic Models (ICML 2022)
·DPM-Solver-v3: Improved Diffusion ODE Solver with Empirical Model Statistics(NeurIPS 2023)
高效训练
·Memory efficient optimizers with 4-bit states
·Training Transformers with 4-bit Integers
·Towards Accelerated Model Training via Bayesian Data Selection
可控生成
·A Closer Look at Parameter-Efficient Tuning in Diffusion Models
·EGSDE: Unpaired Image-to-Image Translation via Energy-Guided Stochastic Differential Equations (NeurIPS 2022)
·Equivariant Energy-Guided SDE for Inverse Molecular Design (ICLR 2023)
多模态训练
·ProlificDreamer: High-Fidelity and Diverse Text-to-3D Generation with Variational Score Distillation(NeurIPS 2023)
·ControlVideo: Conditional Control for One-shot Text-driven Video Editing and Beyond
强化学习
·Contrastive Energy Prediction for Exact Energy-Guided Diffusion Sampling in Offline Reinforcement Learning(ICML 2023)
·Offline reinforcement learning via high-fidelity generative behavior modeling
基础理论
·Improved Techniques for Maximum Likelihood Estimation for Diffusion ODEs(ICML 2023)
·Robust Classification via a Single Diffusion Model
·Diffusion models and semi-supervised learners benefit mutually with few labels(NeurIPS 2023)
上述论文有12篇被机器学习三大顶会接收: ICML、ICLR、NeurIPS,当然这些论文成果不仅停留在paper上,很多成果已被OpenAI ,Apple,stability.ai采用
Vidu核心技术建立在以上顶级基础研究之上
其核心技术U-ViT架构由团队于2022年9月提出,早于Sora采用的DiT架构,是全球首个个Diffusion与Transformer融合的架构
Vidu核心团队成员来自清华大学人工智能研究院,此外汇集了来自阿里、腾讯、字节等知名科技公司的顶尖人才
自从ChatGPT推出以后,OpenAI成为了全球人工智能发展的风向标,GPT3.5,GPT4,sora 三个里程碑影响了无数初创公司,OpenAI也变成了最神秘的人工智能公司,一举一动都受到各方关注,现在vidu的出现刷屏国内外各大媒体平台,一举打破了OpenAI神秘性
⭐星标AI寒武纪,好内容不错过⭐
用你的赞和在看告诉我~
你怎么看?👇👇