Core Viewpoint - The article highlights the significant presence and influence of Chinese companies at the Tokyo Game Show (TGS), showcasing advancements in AI technology and its integration into the gaming industry [1][36]. Group 1: Chinese Companies at TGS - Major Chinese gaming companies such as NetEase, Tencent, and others have established impressive exhibition spaces, attracting numerous players [2][8]. - AI companies are also making their mark at TGS, demonstrating their capabilities and innovations in the gaming sector [8][10]. Group 2: AI Technology Showcase - Alibaba's booth prominently featured its open-source models, including Tongyi Qianwen and Tongyi Wanxiang, offering a range of commercial solutions from IaaS to SaaS [11][12]. - The Model Studio platform and AI development platform PAI were highlighted as part of Alibaba's offerings, indicating a strong push for AI integration in gaming [13][15]. Group 3: 3D Generation Technology - Tencent Cloud emphasized its cloud computing capabilities for game security and operations, while also discussing the potential of mixed reality 3D technology [21][22]. - VAST's Tripo, a leading open-source 3D generation project, is gaining attention from game developers both domestically and internationally [26][27]. Group 4: AI Applications in Gaming - HakkoAI, an AI gaming companion, showcased its ability to understand and interact with various games, outperforming several top general models in specific gaming scenarios [34]. - The integration of AI in gaming is creating new possibilities and enhancing player experiences, indicating a growing trend in the industry [36].

Core Viewpoint - The article discusses the introduction of PhysXNet, the first systematically annotated physical property 3D dataset, which aims to bridge the gap between virtual 3D generation and physical realism [1][3]. Group 1: Introduction of PhysXNet - PhysXNet contains over 26,000 richly annotated 3D objects, covering five core dimensions: physical scale, materials, affordance, kinematic information, and textual descriptions [3][11]. - An extended version, PhysXNet-XL, includes over 6 million programmatically generated 3D objects with physical annotations [12]. Group 2: Current Research Landscape - Existing 3D generation methods primarily focus on geometric structure and texture, neglecting the modeling based on physical properties [2][8]. - The demand for physical modeling, understanding, and reasoning in 3D space is increasing, necessitating a comprehensive physical-based 3D object modeling system [8][9]. Group 3: Data Annotation Process - The team designed a human-in-the-loop annotation process to efficiently collect and annotate physical information [16][19]. - The annotation framework consists of two main phases: initial data collection and determination of kinematic parameters [19]. Group 4: Generation Methodology - PhysXGen is introduced as a novel framework for generating 3D assets with physical properties, utilizing pre-trained 3D priors to achieve efficient training and good generalization [13][26]. - The method synchronously integrates basic physical properties during the generation process, optimizing structural branches for dual objectives [29][30]. Group 5: Experimental Evaluation - The team conducted qualitative and quantitative evaluations of the model, comparing it against a baseline that uses a separate structure to predict physical properties [33][34]. - PhysXGen demonstrated significant performance improvements in generating physical attributes, achieving relative performance gains of 24%, 64%, 28%, and 72% across various dimensions [38]. Group 6: Future Directions - The article emphasizes the importance of addressing key challenges in physical 3D generation tasks and outlines future research directions [43].

Core Insights - The article highlights the significant participation of Chinese companies in CVPR 2025, showcasing their technological advancements and commitment to AI development [4][9][46] - Key trends identified include a focus on multimodal and 3D generation technologies, with Gaussian Splatting emerging as a prominent technique [8][15][17] Group 1: Event Overview - CVPR 2025 has gained increased attention and social engagement, with a record number of Chinese enterprises participating [2][4] - The conference is recognized as a leading event in the field of computer vision, with the acceptance of papers indicating cutting-edge technological trends [12][13] Group 2: Research Trends - Multimodal and 3D generation are highlighted as popular research directions, with Gaussian Splatting being a frequently mentioned keyword in accepted papers [8][15][17] - A total of 2878 papers were analyzed, revealing high-frequency terms such as "Multimodal" (75 occurrences) and "Diffusion Model" (153 occurrences) [16] Group 3: Chinese Companies' Participation - Chinese companies, particularly Tencent, have shown deep involvement, with Tencent alone having over 40 accepted papers across various research areas [33][34] - The participation of Chinese firms in sponsorship and workshops indicates their commitment to the conference and the broader AI landscape [36][38] Group 4: Technological Advancements - Tencent's investment in AI research is substantial, with R&D spending exceeding 70.686 billion RMB in 2024, reflecting a strong commitment to technological innovation [46] - The company has also made significant strides in patent applications, with over 85,000 applications filed globally [46] Group 5: Talent Attraction - The presence of Chinese companies at top conferences serves to attract talent, emphasizing the importance of technical recognition over salary for top-tier professionals [47] - Tencent's diverse application scenarios, including WeChat and gaming, provide a robust ecosystem that supports ongoing technological development [49][50]

Core Insights - The CVPR 2025 conference showcased significant participation from Chinese companies, highlighting their growing influence in the global AI and computer vision landscape [3][7][30] - The conference emphasized advanced topics such as multimodal and 3D generation technologies, with Gaussian Splatting emerging as a key focus area [6][15][17] - The acceptance rate for papers at CVPR 2025 was 22.1%, indicating a competitive environment and increasing recognition for high-quality research [11][13] Group 1: Conference Highlights - The conference received a record number of submissions, with 13,008 valid papers and 2,878 accepted, reflecting a growing interest in cutting-edge research [11] - Key topics included multimodal models, diffusion models, and large language models, with "multimodal" appearing 175 times in accepted paper titles [14] - The integration of computer vision and graphics was noted, with a significant rise in 3D-related research due to advancements in neural rendering [17][18] Group 2: Chinese Companies' Participation - Chinese companies, particularly Tencent, demonstrated strong engagement, with Tencent alone having over 40 accepted papers across various research areas [32] - The participation of Chinese firms in sponsorship and workshops indicates their commitment to advancing technology and attracting talent [34][36] - Tencent's investment in R&D reached approximately 70.686 billion RMB in 2024, showcasing their dedication to AI and technology development [44] Group 3: Talent Acquisition and Development - The conference served as a platform for companies to attract top talent, with Tencent's "Qingyun Plan" offering competitive salaries and career advancement opportunities [50][51] - The focus on technical talent is evident, with 73% of Tencent's workforce in technology roles, emphasizing the importance of skilled personnel in driving innovation [51] - The initiative aims to create a positive cycle where talent is nurtured and retained, contributing to the company's long-term technological advancements [46][48]

Core Insights - VAST has successfully completed a multi-million dollar Pre-A+ funding round led by the Beijing Artificial Intelligence Industry Investment Fund, with participation from Jingya Capital and other investors [1][12] - The company has launched Tripo Studio, the world's first AI-driven all-in-one 3D workspace, and is set to release the new algorithm Tripo 3.0, focusing on the development of the Tripo series of large models and the construction of an ecosystem platform [1][2] - VAST aims to create a comprehensive product system that covers professional (PGC), influencer (PUGC), and general user (UGC) creator profiles, solidifying its global leadership in the 3D generation field [1][3] Funding and Investment - The recent funding round will primarily be invested in the research and development of the Tripo series and the Tripo Studio product [1] - The Beijing Artificial Intelligence Industry Investment Fund and Jingya Capital express confidence in VAST's potential in the 3D model generation sector, highlighting the company's innovative capabilities and market opportunities [11][12] Product Development - VAST has iterated on the Tripo large model series, launching versions from Tripo 1.0 to Tripo 2.5, and has developed widely recognized 3D foundational models [2] - Tripo Studio has received high praise from users, with a 2.5x increase in platform payment rates and an annual recurring revenue (ARR) surpassing $3 million [2] - The company has introduced several innovative features in Tripo Studio, including intelligent part segmentation, magic texture brushes, intelligent low-poly generation, and automatic rigging, significantly enhancing the 3D creation process [4][5][6][8] Market Position and User Engagement - VAST has provided services to over 2 million 3D creators, 20,000 small developers, and 700 large enterprises, generating nearly 30 million models [2] - The company aims to redefine the 3D content creation process, allowing non-professional users to independently complete the entire workflow [9] - VAST collaborates with various industries, including gaming, industrial design, and home 3D printing, to enhance user engagement and creativity in 3D content generation [10] Future Outlook - VAST's CEO emphasizes the shift from merely providing tools to delivering complete solutions that enhance creator control and creativity [11] - The company envisions a future where 3D content creation becomes as ubiquitous and creative as photography, transforming the industry landscape [12]

Core Viewpoint - Step1X-3D is a newly released and open-sourced 3D model with a total parameter count of 4.8 billion, designed to generate high-fidelity and controllable 3D content for various applications including gaming, film, and industrial design [1][3]. Group 1: Data and Algorithm Optimization - Step1X-3D is built on a foundation of over 5 million raw data points, resulting in a training sample library of 2 million high-quality, standardized samples, addressing the industry's data scarcity and quality issues [4]. - The model employs enhanced mesh to SDF conversion techniques, improving the success rate of watertight geometry conversion by 20%, thus enhancing its generalization ability and detail capture [7]. Group 2: 3D Native Generation - The model features a two-stage architecture that decouples geometry and texture representation, ensuring the generated models are structurally reliable and visually accurate, avoiding geometric distortion [10]. - The geometry generation utilizes an innovative mixed VAE-DiT architecture to produce watertight TSDF representations, capturing rich geometric details through techniques like sharp edge sampling [15]. - Texture generation is optimized using a powerful SD-XL model, ensuring vibrant colors and realistic textures that maintain consistency across multiple views, effectively avoiding common distortions and seams [16]. Group 3: Control and Usability - Step1X-3D significantly enhances the controllability and usability of 3D content generation, allowing users to intuitively adjust various attributes such as symmetry and surface details [18][19]. - The architecture's design aligns closely with mainstream 2D generation models, facilitating the integration of established 2D control techniques, thus making the creation process more precise [18]. Group 4: Performance Evaluation - Step1X-3D underwent rigorous quantitative and qualitative assessments, outperforming several mainstream models in key dimensions, particularly achieving the highest CLIP-Score among compared models, indicating strong content and input semantic consistency [23][25]. Group 5: Team and Vision - The development teams, Step1X-3D and LightIllusions, aim to advance AGI and focus on 3D AIGC and spatial intelligence technologies, with a commitment to enhancing 3D content production capabilities and commercializing 3D applications [27].

Core Viewpoint - VAST has launched new 3D generative models, TripoSG and TripoSF, which have set new state-of-the-art (SOTA) benchmarks in the open-source 3D generation field, showcasing significant advancements in quality, detail, and performance [6][8][12]. Group 1: Model Launch and Features - TripoSG is a foundational 3D generative model that has achieved a new SOTA in open-source 3D generation, emphasizing quality, detail, and fidelity [14][16]. - TripoSF, currently in its first phase of open-source release, has proven its capabilities by surpassing existing open-source and closed-source methods, also achieving a new SOTA [8][16]. - VAST plans to continue its open-source initiative for a month, releasing new projects weekly, including various advanced 3D models and techniques [10][66]. Group 2: Technical Innovations - TripoSG incorporates several key design innovations, including the application of a Rectified Flow-based Transformer architecture for 3D shape generation, which offers a more stable and efficient training process compared to traditional diffusion models [21][22]. - The model is the first in the 3D domain to utilize a Mixture of Experts (MoE) Transformer, enhancing feature fusion and allowing for efficient integration of global and local image features [23][24]. - VAST has developed a high-quality Variational Autoencoder (VAE) with innovative geometric supervision, utilizing Signed Distance Functions (SDFs) for improved precision in geometric representation [28][30]. Group 3: Performance Metrics - TripoSG has been evaluated using Normal-FID and other quantitative metrics, demonstrating superior performance in semantic consistency and the ability to accurately reflect the input image's semantic content [34][35]. - TripoSF has achieved approximately 82% reduction in Chamfer Distance and about 88% improvement in F-score across multiple benchmark tests, indicating its high-quality output [57]. Group 4: Future Developments - VAST's upcoming projects include a comprehensive suite of 3D generation technologies, with plans for models focused on 3D component completion and general 3D model binding generation [66][67]. - The final week of the open-source month will feature cutting-edge explorations in 3D generation, including geometric refinement models and interactive sketch-to-3D models [68][69]. Group 5: Industry Impact - VAST is recognized as a leading company in the 3D generative model space, actively contributing to the open-source community and pushing the boundaries of 3D content creation technology [80][87]. - The company aims to democratize 3D content creation, making it accessible to everyone by the end of 2025, aligning with the broader trend of advancing AIGC technologies [85][86].

高校，创新源头。 作者 I 刘博 陈晓 报道 I 投资界PEdaily 一群年轻面孔闯入创投圈。 投资界获悉，3D生成大模型公司影眸科技完成数千万美元 A 轮融资，由美团龙珠、字节 跳动领投，老股东红杉中国种子基金及奇绩创坛持续跟投。令人意外的是，公司团队平均 年龄只有2 4 岁。 这是一个孵化自上海科技大学的创业项目——2 0 2 0年，吴迪、张启煊、张龙文、曾初啸 等人创立影眸科技，团队与上科大共同提出的可控 3D 原生 DiT生成框架 CLAY 与 3D 服装生成框架 Dr e ssCod e，均获计算机图形学顶会 ACM SIGGRAPH 2 0 24 最佳论文提 名，被认为是新一代 3D 生成基础框架。 "现在我们经常跑上科大蹲项目。"此前一家知名早期投资机构的分享引起我们的注意。 不同于清华、上海交大、哈工大等传统名校，上科大乍听略显陌生，成立仅仅十余年，殊 不知已经累计孵化4 0多家科创企业。如此一幕，堪称中国科技成果转化大潮的一缕写 照。 团队平均24岁 刚刚，美团字节联手投了 影眸科技的故事，始于上科大的一间实验室。 出生于19 9 7年，吴迪在20 1 5年进入上科大学习，是该校招的第二届 ...