news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Smart Things
Author: Cheng Qian
Edit Heart

After 10 years, the 15 "Four Little Dragons of CV (Computer Vision)" are creating huge waves in the big model entrepreneurial boom.

In 2014-2015, SenseTime, Megvii Technology, Yitu Technology, and CloudWalk Technology emerged in the first wave of AI entrepreneurship and were collectively known as the "Four Little Dragons of CV"; nearly 10 years later, employees of the "Four Little Dragons of CV" once again joined the second wave of AI entrepreneurship.

According to Zhidongxi statistics, since ChatGPT swept the world, there have been 10,000 employees from the "Four Little Dragons of CV" who have devoted themselves to generative AI entrepreneurship.15 people, founded9Start-up business.

Among them, two unicorns were born:MiniMax, Dark Side of the Moon, and has become the "top stream" of large-scale model entrepreneurship in China.

9 startups coverLarge model full industry chain, entrepreneurial directions include AI Agent, general large model, AI Infra, AI application, etc.

Among the “Four Little Dragons of CV”,SenseTimeThe employees have the most9-digit，Megvii TechnologyNext is4-digit，2nd place, Yitu Technology, which was not found among founders of large model startupsCloudWalk TechnologyThe figure of a former employee.

These startups are not enough to describe as having penetrated the entire domestic large model industry chain, attracting a lot of money, and having a luxurious lineup of founders. These industry giants who are deeply involved in the current AI new entrepreneurial wave are competing with the "Four Little Dragons of CV" and becoming potential players in the large model industry.

▲Employees of the “Four Little Dragons of CV” resigned and started generative AI startups (data compiled from public information, please point out any errors and omissions)

Today, the new four big model dragons have not yet been determined, but the big model startups incubated by former employees of the "CV Four Little Dragons" are crossing the cycle to continue writing the AI ​​myth.

1. Former SenseTime employees are keen to form groups to start businesses and compete with their former employers

In 2021, SenseTime fired the first shot in the listing of CV unicorns. Focusing on original technology research in computer vision and deep learning, it has become the largest AI algorithm provider in Asia.

Today, big models are advancing rapidly, and employees who have left SenseTime have become the largest number of those entering the big model startup industry among the "Four Little Dragons of CV", including 8 employees and 5 startups, and 4 of them are startups formed by former SenseTime employees. Their main focus covers general big models, multimodal big models and AI applications.

1. MiniMax: The founder of MiniMax has the highest content of SenseTime, and has become a large-scale unicorn

The valuation has exceeded $2.5 billionMiniMaxAlmost all of the resumes of founders to senior executives include experience at SenseTime.

Co-founder of MiniMaxYan JunjieHe is the former vice president of SenseTime and the head of general intelligent technology. He was responsible for building the tool chain and underlying algorithms for deep learning at SenseTime.

▲MiniMax co-founder Yan Junjie

Early co-founderZhou Yucong, Yun YeyiHe also worked at SenseTime. Yun Yeyi was once the strategic director of the CEO office of SenseTime and the assistant of SenseTime co-founder Xu Li. Zhou Yucong was in charge of the algorithm team at SenseTime. When he was a student, he won the first place in the ASC15 finals of the World University Supercomputer Competition and the International Supercomputer Conference ISC17. Later, he participated in the International Conference on Computer Vision ICCV 2019 Challenge and won the iQIYI-Large track championship.

MiniMax, which was founded in December 2021, developed a large model with 30 billion parameters before the release of ChatGPT. In October 2022, it released the conversational interactive intelligent product Glow. It has currently developed a trillion-parameter MoE text large model, voice large model and image large model, and based on this, it has created applications such as the AI ​​companion product Xingye and the smart assistant Conch AI.

In terms of financing, MiniMax has completed four rounds of financing, of which two have disclosed specific amounts, namely a round A financing of over US$250 million in June last year and a strategic financing of US$600 million in March this year. Investors include Tencent Investment, Alibaba, etc.

2. Right Brain Technology: It used to focus on video generation at SenseTime. Its first product had over one million users within two months of its launch.

Right Brain TechnologyCo-founderShi Jie, formerly worked as a researcher at SenseTime, responsible for the research of diffusion video generation models.

▲Shi Jie, co-founder of Right Brain Technology (far right) (Photo source: PuYue China)

The company was founded in September 2022 and focuses on the research and development of AI image and video generation. In 2023, the company's first video generation product Vega started internal testing. At that time, there were almost no video generation-related products in China, so the number of users of this product increased rapidly. Less than two months after its launch, the number of Vega users exceeded one million.

At present, its two main products, AI video tool Video Studio and image creation platform Vega AI, have been released.

The company has received three rounds of financing, two of which amounted to tens of millions of yuan. Investors include Qiji Chuangtan, Lightspeed Photosynthesis, and ZhenFund.

3. VAST: A continuation of AIGC 3D, and once participated in the entrepreneurship of MiniMax

VASTFounder and CEOSong YachenIn 2019, he mainly worked on AIGC 3D at SenseTime, which was consistent with his subsequent entrepreneurial direction. In 2021, Song Yachen joined the AI ​​Game Division of SenseTime, and then he participated in the establishment of MiniMax as No. 001.

Because he came into contact with 3D content, generative AI-related content and friends during work, he saw the prospects of creating 3D content, and the establishment of VAST was a natural outcome.

Another co-founderLiang DingAlso coming from a technical background, he served as the initiator of the general vision and language big model and the person in charge of the vertical language big model at SenseTime. During his tenure, he led the launch of the country's first mobile phone face unlocking, face door locks, and city-level subway face-scanning rides.

▲VAST co-founder Liang Ding (Source: LinkedIn)

VAST was founded in November 2022. In December last year, its Tripo, a 3D model with billions of parameters, was launched, generating more than 2 million 3D models in more than 100 days. Currently, the startup has more than 200 B-side customers and has developed native AI hit games based on its platform, with more than 100,000 users in a week.

However, the startup has not yet disclosed its financing situation.

4. Yantu Intelligence: Participated in the design of the first batch of police dynamic face recognition systems and customized Agents for enterprises

Word and Picture IntelligenceThe founder isLuo YuchenHe joined SenseTime in 2017 to work on security-related projects, and transferred to the EIG department in 2019, where he participated in algorithm public relations under small data samples, long-tail large-scale algorithm system delivery, and multimodal algorithm comprehensive review system. During his tenure, he participated in the design of China's first batch of dynamic face recognition systems for police use.

▲Luo Yuchen, founder of Yantu Intelligence (Photo source: Qichacha)

Co-founder and head of product at the companyWang ZekaiHe is also an employee of SenseTime and was a senior product expert.

▲Wang Zekai, co-founder and product manager of Yantu Intelligence (Photo source: Foshan University of Science and Technology Shenzhen Alumni Association)

Yantu Intelligence was established in March last year, mainly to provide customized AI Agent autonomous intelligent body services for enterprises based on large language models. The company's products include the enterprise intelligent body platform "Yanxin" and the knowledge workbench "Yanzhi". Yanxin helps enterprises create intelligent bodies that can answer questions in customer service, operations, digital humans and other scenarios, and the Yanzhi workbench can create complex task intelligent bodies.

5. Wuwen Xinqiong: Building a 20,000-chip GPU computing platform at SenseTime, the startup has received investment from Tencent and Baidu

The only AI Infra startup founded by SenseTime employees isNo Question Core Dome。

Co-founder and CTO of Wuwen XinqiongYan ShengenHe served as the executive research director of the Data and Computing Platform Department at SenseTime, helping SenseTime build a large-scale high-performance AI computing platform with 20,000 GPUs. He also presided over the development of multiple deep learning system software and led a team of 200 people to build the Shanghai AI supercomputer prototype project in three years.

▲ Yan Shengen, co-founder and CTO of Wuwen Xinqiong (Photo source: China Optics Valley International Artificial Intelligence Industry Forum)

Founded in May last year, Wuwen Xinqiong has realized the Infini-AI large model development and service cloud platform in just over a year. It supports large model players such as Zhipu AI, Dark Side of the Moon, and Shengshu Technology to stably use heterogeneous computing power, and makes the computing power utilization rate of the Qianka heterogeneous hybrid training cluster reach up to 97.6%, supporting large model hybrid training including six heterogeneous chips including AMD, Huawei Ascend, Tianshu Zhixin, Muxi, Moore Thread, and NVIDIA.

This startup has a luxurious lineup of investors and has completed two rounds of financing but has not disclosed the specific amount. Leading players in large models such as Baidu, Tencent and Zhipu AI, as well as top investment institutions such as Sequoia China, GSR Ventures, Xuhui Capital and Infinite Fund are among its investors.

These five startups have different directions, ranging from language, video, and 3D in the general large model to AI applications and AI Infra. Another notable feature is that most of them started their own businesses with a few colleagues from SenseTime.

SenseTime, which fired the first shot of CV unicorn listing, has become the "Huangpu Military Academy" in the field of computer vision. Now SenseTime has also become a leading player in the field of large models. Earlier this month, SenseTime released the RiRiXin 5.5 and the 5o version that can perform streaming multimodal interaction, and many of its evaluations have exceeded GPT-4o.

2. Three employees of Megvii Technology left to start their own business, and a popular unicorn was born

Megvii TechnologyAlthough it is not as popular as SenseTime in the era of big models, many of its employees have left to start generative AI startups. Many of Megvii’s former employees have also joined generative AI startups, including three former employees, covering three startups.

The most eye-catching ones must be the current hottest big model company, Dark Side of the Moon, as well as startups focusing on AI applications and computing power.

1. Dzine AI: Worked at Adobe, Megvii, and Tencent, currently focusing on AI photo editing

Dzine AIThe founder is the head of Megvii Technology's American Research InstituteWang JueBefore starting his own business, Wang Jue worked at Adobe for nine years. In 2017, he joined Megvii Technology as the director of the US laboratory. Many cutting-edge mobile photography solutions he led to develop at Megvii Technology have been shipped in tens of millions of Android smartphones. He then joined Tencent.

▲Wang Jue, founder and president of Dzine AI (Photo source: Megvii Technology)

Wang Jue has published more than 200 research papers in top academic journals and conferences in the fields of computer vision, computer graphics, machine learning, and human-computer interaction, and holds more than 80 international patents.

After leaving Tencent in November 2022, Wang Jue co-founded Dzine AI, an AI application startup. Currently, the company has launched Dzine, an AI photo editing tool that helps designers adjust image styles, generate fill-in images, or modify them based on AI capabilities.

2. The Dark Side of the Moon: A researcher at Megvii was responsible for the mass production of algorithms, and the startup’s valuation has exceeded $3 billion

Dark Side of the MoonIt is the second large-scale unicorn founded by employees of the “Four Little Dragons of CV” who resigned to start their own businesses.

Co-founder of Dark Side of the Moon, Head of Algorithm TeamZhou XinyuHe was in charge of algorithm mass production at Megvii Research Institute. During his research on mobile models, he wrote the paper "ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices" as a co-first author and won the CVPR bid. ShuffleNet later influenced various mobile phone millisecond-level face unlocking technologies, including Apple's 3D face unlocking.

▲Zhou Xinyu, co-founder of Dark Side of the Moon and head of the algorithm team (Photo source: Tsinghua Alumni Association)

After co-founding Dark Side of the Moon, Zhou Xinyu served as the head of the algorithm team. Dark Side of the Moon is a general-purpose large model in China, and has now released its first conversation product, Kimi Chat. In March this year, Kimi Chat announced that it could support ultra-long conversation contexts of about 2 million words, which led to a surge in the number of users and caused multiple downtimes that became a hot topic.

Dark Side of the Moon became a unicorn less than a year after its establishment. It is now rumored to have completed three rounds of financing, with a disclosed financing amount of over US$1.3 billion and a valuation of over US$3 billion. Star players such as Sequoia China, Xiaohongshu, Meituan, Alibaba, and Tencent are among its investors.

3. Yixin Technology: Two former employees jointly start a business, focusing on AI computing power

Yixin TechnologyCo-FounderWu WenhaoHe used to be the senior vice president of the cloud business unit at Megvii Technology, responsible for 3D+AI intelligent perception solutions, which were mainly used in areas such as mobile phone face recognition.

▲Wu Wenhao, co-founder of Yixin Technology (Photo source: Maimai)

Co-founder of Yixin TechnologyWang ZhichengHe has also worked as a senior researcher at Megvii Technology and has won multiple international championships in face detection competitions on behalf of Megvii Technology. He is currently the head of algorithms at Yixin Technology.

▲Wang Zhicheng, co-founder of Yixin Technology

Public information from venture capital data service provider IT Juzi shows that Yixin Technology's main direction is the research and development and optimization of conversational products based on large language models, such as large model training and fine-tuning, accelerated optimization, etc.; AIGC product research and development, such as multi-modal pre-training and fine-tuning of Vincent videos, etc.

Tianyancha shows that in January this year, Yixin Technology has received angel round financing from Qifu Capital and Cathay Ventures. At the same time, in April this year, Cathay Ventures publicly disclosed that Yixin Technology is a representative case of its investment layout in computing power clusters in 2023.

The three startups focus on AI applications, general large models, and computing power.

Megvii Technology has also built a complete business layout for the era of big models. Megvii has developed its own AI productivity platform Brain++, including the deep learning framework MegEngine (Megvii Tianyuan), the deep learning cloud computing platform MegCompute, and the data management platform MegData, covering all aspects from AI production (outputting algorithm models) to application (realizing algorithm engineering packaging).

3. Former Yitu employees teamed up to win tens of millions of yuan in financing.CloudWalk Technology employees

Compared with SenseTime and Megvii Technology, the number of former employees from the other two “CV Four Little Dragons” entering generative AI startups is relatively small.

1. Lanma Technology: Former “veteran” employees jointly start a business and have received tens of millions of yuan in financing

There are two people who left Yitu Technology to start their own business.Lanma TechnologyFounder and CEOZhou Jianand co-founderZhou Yuanjian, they are respectively the 10th and 6th employees of Yitu Technology.

▲Zhou Jian, founder and CEO of Lanma Technology (Source: Lanma Technology)

Zhou Jian participated in the planning and construction of Yitu's technology and engineering team from 0 to 1. He was responsible for the engineering implementation of AI algorithms, designed and delivered a billion-level image retrieval system, etc.; Zhou Yuanjian was responsible for Yitu Technology's product research and development in CV, Speech, Big Data and other fields.

▲Zhou Yuanjian, co-founder of Lanma Technology (Source: Lanma Technology)

The company they jointly founded, Lanma Technology, is an enterprise-level AI Agent platform company based on a large language model. It has already released the enterprise-level AI Agent platform AskXBOT. Enterprises can use this platform to quickly build commonly used agents for various segmented applications such as HR recruitment, document question and answer, and data query.

In August last year, the company completed a tens of millions of yuan Series A financing round led by IDG.

2. Kunlun Wanwei: Former CTO joins the big model game, and the team has released an open source multimodal big model

In addition to starting their own businesses, Yitu Technology also has former employees who choose to join the big model track.

Last March,Yan Shuicheng, former CTO of Yitu TechnologyHe joined the AI ​​Research Institute as a visiting chief scientist. Yan Shuicheng then joined Kunlun Wanwei in September last year as co-CEO of Tiangong Intelligent Technology and dean of the 2050 Global Research Institute.

▲Yan Shuicheng, Co-CEO of Tiangong Intelligent Technology and Dean of 2050 Global Research Institute (Photo source: Zhiyuan Community)

In April this year, Yan Shuicheng's team jointly released and open-sourced the Vitron general pixel-level visual multimodal large model with the National University of Singapore and Nanyang Technological University. Vitron solves the image/video model split problem that has long plagued the large language model industry, including comprehensive understanding, generation, segmentation and editing of static images and dynamic video content.

At present, Tiangong App, a product of Kunlun Wanwei Tiangong Intelligent, has integrated core functions such as AI search, AI reading, and AI creation.

Its former employer, Yitu Technology, has targeted the field of smart security, in which it excels. In June this year, it released the multimodal Yitu Tianwen Multimodal Big Model 4.0, which is the first practical and commercially available multimodal big model in the field of smart security. It can cold-start new algorithms for very few samples within 1 minute, complete online labeling training within 1 hour, and quickly deploy it online within 1 day.

3. CloudWalk Technology: No former employees have publicly started businesses yet, and a complete business line of large models has been built

The last of the "Four Little Dragons of CV"——CloudWalk TechnologyJudging from the public information, there is no new trend of former employees joining the generative AI startup industry.

In May 2022, CloudWalk Technology was successfully listed on the Science and Technology Innovation Board of the Shanghai Stock Exchange, becoming the "first AI platform stock" on the Science and Technology Innovation Board. In May last year, it also officially entered the big model market, and the multimodal big model Congrong Big Model was released, which has now been iterated to version 1.5. At that time, this 13 billion parameter scale industry big model distilled from the 100 billion big model exceeded ChatGPT in 44 of the 52 test indicators and GPT-4 in 29.

At present, Yitu Technology has a complete layout from base model, application ecology to intelligent computing center, including dozens of large industry models, and has developed multiple pan-AI intelligent applications such as DataGPT, intelligent customer service, and AI mouse.

Conclusion: The next wave has arrived! Continuing the AI ​​2.0 entrepreneurial boom

The "CV Four Little Dragons" that once dominated the AI ​​wave have experienced the AI ​​bubble dissipation, and are still facing difficulties in their transformation despite losses. Now that the big model track is surging, these employees who left the "CV Four Little Dragons" have become participants in the new wave of AI.

From general big models, AI computing power to AI applications, these big names can be found in the entire big model industry chain. These entrants who have deeply participated in the AI ​​wave and experienced several twists and turns may be able to tap into unique market advantages in new AI opportunities.