


Apple recently announced that it will launch its first virtual head-mounted display device, Vision Pro, on February 2. This XR device is expected to lead the rapid development of the next generation of terminals. With the popularization of virtual display devices, digital interaction will move from two-dimensional to three-dimensional, and three-dimensional models and three-dimensional animation will become the mainstream content forms in the future. Multi-dimensional immersive interaction will also become a trend under the trend of virtual and real integration.
However, from the perspective of data scale, the data accumulation in the content industry at this stage is still mainly 2D images and flat videos, while the data foundation of 3D models, 4D animations and other data is relatively weak. Among them, 4D animation introduces time sequences on the basis of traditional 3D models, that is, 3D models that change over time, which can present dynamic three-dimensional effects. It is widely used in game animation, movie special effects, virtual reality and other fields, but it is also currently The most difficult link in content ecological development.
Therefore, for the upcoming multi-dimensional immersive experience, building three-dimensional digital content will become an important basic work.
Facing this cutting-edge field, Tsinghua University’s entrepreneurial team Shengshu Technology has carried out a series of research and product development, and recently launched the world’s first 4D based on “skeletal animation” in conjunction with Tsinghua University, Tongji University and other universities The animation generation framework "AnimatableDreamer" can directly convert 2D video materials into dynamic three-dimensional models (i.e. 4D animation) with one click. It supports automatic extraction of skeletal movements, one-click conversion of animation effects, and personalized character generation through text input. .
Paper address: https://arxiv.org/pdf/2312.03795.pdf
Project Address: https://animatabledreamer.github.io/
Paper title: AnimatableDreamer: Text-Guided Non-rigid 3D Model Generation and Reconstruction with Canonical Score Distillation

New content production method
Subversion of the three-dimensional animation development process
Past Industry Producing three-dimensional animation requires complex processes such as 3D modeling, texture rendering, bone binding, and animation production. It requires the participation of professionals such as modelers and animators to complete, which is low-efficiency and high-cost. According to statistics, the production cycle of a single static 3D model is from hours to days, and the cost can be as high as thousands of dollars. The cost of dynamic processing will be even higher.
As can be seen from the official demo video, upload a 2D real-life video of a squirrel, and enter the text description of "A squirrel with red sweater" to create the original real-life squirrel with one click Convert to animation style, and perfectly retain the action postures. Synchronization can also generate a three-dimensional dynamic model from a 360-degree perspective. By switching text descriptions, you can customize the character and switch the squirrel to different cartoon images such as fox and Squirtle. .
"AnimatableDreamer" can automatically extract the skeletal movements of target objects (characters, animals, etc.) in the video, and then convert the object into any skeletal animation model through text description. The entire process is not limited by templates, supports any video length and any type of action, has a high degree of temporal consistency and multi-view consistency, and the exported dynamic three-dimensional model can be rendered in any 3D environment.
For post-production in film, television, animation and other fields, more editing is usually required for animation. "AnimatableDreamer" also supports the replacement or editing of animation files for models that have completed bone binding, achieving a higher degree of freedom. . When related technologies are gradually implemented in the future, the 3D modeling and animation production processes for game development, film and television animation and other scenes are expected to become more efficient.
Facing the future
It is expected to form a new content ecology
In terms of implementation principles, the research team innovatively proposed Canonical Score Distillation (CSD) renders and denoises 3D models of different frames and different perspectives in the camera space that changes over time, and uniformly returns the gradients to different Distillation is performed in the canonical space shared by the camera space, and the dimensionality of 4D generation is reduced to 3D, that is, the 4D generation problem is simplified into generation in 3D space.
"AnimatableDreamer" can extract joint movements directly from videos. By decoupling the object's model and motion, the generated model has a high degree of temporal consistency and geometric rationality, is not limited by video length, and can be effectively Minimally eliminate issues such as shape breakage, flickering, and multi-view inconsistencies.
In scenes with limited viewing angles and large movements, due to the introduction of prior knowledge of the Diffusion model, even if the input video does not cover the complete object, "AnimatableDreamer" can automatically complete the picture information and achieve better generation quality. .
It can be said that the proposal of "AnimatableDreamer" directly connects text to 4D skeletal animation generation, modeling, mapping, and bone binding , action-driven in one go! Input a natural language description and automatically output a three-dimensional animation video. No professional knowledge is required. Ordinary people can get started and easily customize animation content.
Work based on "AnimatableDreamer" will greatly reduce the difficulty of producing 3D and 4D digital content, enrich the interactive experience, and enable everyone to generate and edit creative content, which will give birth to the 3D era. Create a new content entertainment and content consumption model.
Imagine that in the future virtual world, users can quickly build customized digital spaces and create personalized interactive experiences. For example:
Each character image in the digital space can be generated at will, such as dressing children in Superman clothes, switching to holiday-themed costumes for Halloween, etc.;
Users who have pets can cartoonize their pets, for example, into a virtual Mickey Mouse image. Pets' daily life is just like cartoons, and the daily interaction between owners and pets will become interesting;
The ways of interaction between people will also become richer, and you can have sex anytime, anywhere. For a theme party, you can generate the desired party environment, character costumes, etc. in real time.
As a start-up company established less than a year ago, the Shengshu Technology team has long been committed to the fields of multi-modal large models such as images, 3D, and video, and released 3D assets in September The creation tool VoxCraft, officially launched on Discord, supports text and image guidance, minute-level creation of 3D models, custom replacement of 3D textures and other functions, empowering the 3D modeling process of game development, film and television animation and other scenes. The 4D skeletal animation generation launched this time is another new exploration work of Shengshu Technology, which will be integrated into VoxCraft products in the future.
VoxCraft tool address: https://voxcraft.ai/
The arrival of Apple Vision Pro is not only an important innovation at the hardware device level, but will also start a revolution in content and experience. prelude. The innovative capabilities of generative AI such as 4D animation generation will not only bring better visual presentation, but will also open up multi-dimensional digital experiences in new ways, bringing more possibilities to the next generation of human-computer interaction.
The above is the detailed content of A Tsinghua-based start-up company released the world's first 4D skeletal animation framework, which can convert real scenes into animations with one click and can generate personalized characters.. For more information, please follow other related articles on the PHP Chinese website!

你可能听过以下犀利的观点:1.跟着NVIDIA的技术路线,可能永远也追不上NVIDIA的脚步。2.DSA或许有机会追赶上NVIDIA,但目前的状况是DSA濒临消亡,看不到任何希望另一方面,我们都知道现在大模型正处于风口位置,业界很多人想做大模型芯片,也有很多人想投大模型芯片。但是,大模型芯片的设计关键在哪,大带宽大内存的重要性好像大家都知道,但做出来的芯片跟NVIDIA相比,又有何不同?带着问题,本文尝试给大家一点启发。纯粹以观点为主的文章往往显得形式主义,我们可以通过一个架构的例子来说明Sam

2021年9月25日,阿里云发布了开源项目通义千问140亿参数模型Qwen-14B以及其对话模型Qwen-14B-Chat,并且可以免费商用。Qwen-14B在多个权威评测中表现出色,超过了同等规模的模型,甚至有些指标接近Llama2-70B。此前,阿里云还开源了70亿参数模型Qwen-7B,仅一个多月的时间下载量就突破了100万,成为开源社区的热门项目Qwen-14B是一款支持多种语言的高性能开源模型,相比同类模型使用了更多的高质量数据,整体训练数据超过3万亿Token,使得模型具备更强大的推

在法国巴黎举行了国际计算机视觉大会ICCV(InternationalConferenceonComputerVision)本周开幕作为全球计算机视觉领域顶级的学术会议,ICCV每两年召开一次。ICCV的热度一直以来都与CVPR不相上下,屡创新高在今天的开幕式上,ICCV官方公布了今年的论文数据:本届ICCV共有8068篇投稿,其中有2160篇被接收,录用率为26.8%,略高于上一届ICCV2021的录用率25.9%在论文主题方面,官方也公布了相关数据:多视角和传感器的3D技术热度最高在今天的开

随着智慧司法的兴起,智能化方法驱动的智能法律系统有望惠及不同群体。例如,为法律专业人员减轻文书工作,为普通民众提供法律咨询服务,为法学学生提供学习和考试辅导。由于法律知识的独特性和司法任务的多样性,此前的智慧司法研究方面主要着眼于为特定任务设计自动化算法,难以满足对司法领域提供支撑性服务的需求,离应用落地有不小的距离。而大型语言模型(LLMs)在不同的传统任务上展示出强大的能力,为智能法律系统的进一步发展带来希望。近日,复旦大学数据智能与社会计算实验室(FudanDISC)发布大语言模型驱动的中

8月31日,文心一言首次向全社会全面开放。用户可以在应用商店下载“文心一言APP”或登录“文心一言官网”(https://yiyan.baidu.com)进行体验据报道,百度计划推出一系列经过全新重构的AI原生应用,以便让用户充分体验生成式AI的理解、生成、逻辑和记忆等四大核心能力今年3月16日,文心一言开启邀测。作为全球大厂中首个发布的生成式AI产品,文心一言的基础模型文心大模型早在2019年就在国内率先发布,近期升级的文心大模型3.5也持续在十余个国内外权威测评中位居第一。李彦宏表示,当文心

保险行业对于社会民生和国民经济的重要性不言而喻。作为风险管理工具,保险为人民群众提供保障和福利,推动经济的稳定和可持续发展。在新的时代背景下,保险行业面临着新的机遇和挑战,需要不断创新和转型,以适应社会需求的变化和经济结构的调整近年来,中国的保险科技蓬勃发展。通过创新的商业模式和先进的技术手段,积极推动保险行业实现数字化和智能化转型。保险科技的目标是提升保险服务的便利性、个性化和智能化水平,以前所未有的速度改变传统保险业的面貌。这一发展趋势为保险行业注入了新的活力,使保险产品更贴近人民群众的实际

不得不说,Llama2的「二创」项目越来越硬核、有趣了。自Meta发布开源大模型Llama2以来,围绕着该模型的「二创」项目便多了起来。此前7月,特斯拉前AI总监、重回OpenAI的AndrejKarpathy利用周末时间,做了一个关于Llama2的有趣项目llama2.c,让用户在PyTorch中训练一个babyLlama2模型,然后使用近500行纯C、无任何依赖性的文件进行推理。今天,在Karpathyllama2.c项目的基础上,又有开发者创建了一个启动Llama2的演示操作系统,以及一个

杭州第19届亚运会不仅是国际顶级体育盛会,更是一场精彩绝伦的中国科技盛宴。本届亚运会中,快手StreamLake与杭州电信深度合作,联合打造智慧观赛新体验,在击剑赛事的转播中,全面应用了快手StreamLake六自由度技术,其中“子弹时间”也是首次应用于击剑项目国际顶级赛事。中国电信杭州分公司智能亚运专班组长芮杰表示,依托快手StreamLake自研的4K3D虚拟运镜视频技术和中国电信5G/全光网,通过赛场内部署的4K专业摄像机阵列实时采集的高清竞赛视频,


Hot AI Tools

Undresser.AI Undress
AI-powered app for creating realistic nude photos

AI Clothes Remover
Online AI tool for removing clothes from photos.

Undress AI Tool
Undress images for free

Clothoff.io
AI clothes remover

AI Hentai Generator
Generate AI Hentai for free.

Hot Article

Hot Tools

Safe Exam Browser
Safe Exam Browser is a secure browser environment for taking online exams securely. This software turns any computer into a secure workstation. It controls access to any utility and prevents students from using unauthorized resources.

PhpStorm Mac version
The latest (2018.2.1) professional PHP integrated development tool

SublimeText3 Chinese version
Chinese version, very easy to use

MinGW - Minimalist GNU for Windows
This project is in the process of being migrated to osdn.net/projects/mingw, you can continue to follow us there. MinGW: A native Windows port of the GNU Compiler Collection (GCC), freely distributable import libraries and header files for building native Windows applications; includes extensions to the MSVC runtime to support C99 functionality. All MinGW software can run on 64-bit Windows platforms.

Dreamweaver CS6
Visual web development tools
