Subscribe

Analysis EO
Analysis · 2
report
Analysis EO
May 2, 2020 09:05 pm ·

Unveiling the ‘Mystery’ of Megvii Part 1

Key investment thesis Attractive secular trends. Megvii occupies a unique place within the computer vision domain and benefits from increasing awareness about leveraging artificial intelligence (AI) and the Internet of Things (IoT) in managing business and public management from the government and business side.  Clear long-term strategic map. With the launch of open platform Face++ and open-source deep learning framework MegEngine, we see the company’s ambition in competing with large tech companies as well as grasping the core competence of AI. Unique customer proposition. The full-stack solutions that contain the IoT layout of cloud-edge-devices, platform software systems and applications allow the company to reduce friction around today’s AI, big data and cloud technologies.   Key investment risks Megvii’s cost structure for its SaaS and IoT solutions remains debatable. The business is a bit far away from being a classic SaaS and more like driven by software + services/consulting.  Services revenue weighs on the gross margins. The services business is naturally not as scalable and operates with low margins. Megvii’s business builds on its IoT solutions, contracting with government agencies (more precisely, system integrators). Only with better margins can the business scale faster – not the other way around. Here, the income statements tell the story already. Data ethics problems. Megvii’s services and products have been interacting with a significant amount of sensitive information. It has an additional responsibility to protect its from information leaks and hacks, as well as a duty to deal with regulatory/compliance issues.  Business overview Megvii is the first so-called artificial intelligence unicorn to try to list on the public market in China. The company provides full-stack solutions that encompass algorithms, software and IoT devices to its customers. In this article, we focus on its business model, revenue potential and competitive risks.  Megvii started in a niche area within the face recognition domain. It then launched the first computer vision (CV) open platform – Face++ – of the company in Oct 2012, one year after its inception.  Megvii derives its revenue through two standard pricing models (SaaS subscription and professional services) and one special business that it calls ‘personal devices.’ Like other Software-as-a-Service (SaaS) companies, Megvii charges by a pay-per-use model for its Face ID (a cloud-based identity authentication product) and Face++ under the SaaS business.  Megvii generates revenues for professional services primarily in two segments and charges on a project basis: government and commercial. Worth noting is that a significant portion of Megvii’s direct customers are system integrators, which provide various types of assistance in project implementation, and are not the end users. 93% of revenue in the first half of 2019 came from its city IoT solutions, whereas the rest was generated from its commercial segment supply chain IoT (retail and logistics).  Megvii’s personal devices business accounted for 9% of revenue in 2019H1 Megvii productized the business and earned CNY 5.9 million in 2017, one year earlier than its SaaS offerings. The gross profit saw a sudden plummet in 2019H1 due to the firm delivering more camera modules. Arcsoft (688088:SH), Nuance Communications (NUAN:NASDAQ) and Gracenote, among others, are competing in the niche worldwide. Arcsoft is now trading on China’s new Star market at a 122x earning level with a valuation of CNY 25 billion as of April 24. Arcsoft provides single/dual camera solutions for smartphone manufacturers, which fit various hardware configurations. It has achieved CNY 438 million revenue from its smartphone business, with a 94% gross margin. While Arcsoft’s revenue and gross margin were on a steady growth stage from 2016 to 2018, Megvii saw a meaningful dip in 2019H1, as mentioned above. As a relatively new player in the market, it is reasonable that Megvii has entered into a new stage in delivering products. In essence, it has to work more closely with camera module manufacturers, camera sensor manufacturers and System on a Chip (SoC) platforms. Megvii might need to switch to a different type of camera module for different types of mobile phones to drive the development of camera technologies fast. As a result, the cost of sales surged 17x to CNY 18 million in 2019H1 from one year before. The cost of sales is attributed to two elements: 1) hardware and 2) project outsourcing or technologies services. Considering Megvii’s own claims, we think the previous assumption is reasonable. In essence, the gross margin dip will jump back to normal levels when Megvii reaches the critical mass of producing deliverable camera modules. Megvii’s SaaS offerings, however, are far beyond the SaaS concept that Wall Street and VCs admire  When it comes to Megvii, its SaaS business contains a data source cost, which is unusual for an SaaS offering. China, famous for its extensive public security market (USD 80 billion) as well as significant spending on tech-enabled surveillance (USD 30 billion), is also well-known for its abundant pool of data. But data alone is not enough for building AI software – data must first be labelled, which requires much labor.  Megvii engages third-party data sources for its Face ID product development, which represented 7% and 58% of revenue and total cost of sales in 2019H1, respectively. Data from 2016 to 2019 shows that data source cost as of the total cost of sales reached a plateau of 58% after 2017.  The current machine learning technology is still in an early stage. Thus companies researching it are running algorithms on various datasets to train their models, which means large overheads for purchasing these data sources.  As such, AI businesses are very different from software businesses, whether it is in terms of cost structure or competitive advantages.  The gross margins are lower. An SaaS business means a high gross margin (60% – 80%), with the cost of sales mainly attributed to cloud services. SaaS stocks maintain their gross margins, ranging from 70% to 85%, with a median margin staying at 72% and averaging at 70%, as of April 22, from the 63 SaaS stocks selected. Compared with these mature firms, Megvii’s gross margin shows a continuous growing trend from 2016 to 2019H1. It even reached 87% in 2019H1, a 106% increase compared with 2016. But we do not think this will continue in the near term, mainly from the observation that data source costs will remain a significant part of revenue.  There’s a long tail problem. A study conducted by Arun Chaganty at Eloquent Labs (quoted by a 16z) that researched questions submitted to a chatbot in the customer support space shows a diminishing marginal value of data. When 40% of queries have been collected, there is no advantage to collecting more in this case. AI startups have to devote more time and resources to deal with noisy, unstructured data. They also need to think through edge case difficulties, which is a tough job that traditional software companies won’t face when they build and deploy their products for their earliest customer cohorts. Megvii’s whole R&D special expenditure – including data obtaining and labeling – represented 14.9%, 11.4%, 6.1% and 6.5% of total revenues in 2016, 2017, 2018 and 2019H1, respectively.  AI is early in its development, and has been witnessing sharp declines in both AI training costs and inference costs. In 2018, the cost to train a neural network like ResNet-50 was USD 358. In 2019, it dropped to around USD 20. The cost of inference has dropped meanwhile. The cost to perform inference on 1 million images went from around USD 15 at the end of 2017 to USD 2 in earlier 2019.   AI companies have been improving computing ability at five times the rate of Moore’s law. No wonder there is a massive hype around AI research and funding activities. But in practice, we found that AI businesses have limits in how their value can be built up and accessed, rented or sold – they are not like those 'builds once / sell many times' software model. It adds pressure to margins, as well as defensibilities.  This article is part I of our analysis on Megvii. Please continue to part II.

Analysis EO
Analysis · 2
report
Analysis EO
May 2, 2020 09:05 pm ·

Unveiling the ‘Mystery’ of Megvii Part 2

This article is part II of our analysis on Megvii, check out part I before you read.  The service business: an albatross around the neck – and a future chokepoint  AI is creating a new type of business that contains elements of both software and services.  Some very successful SaaS products attract customers and keep growing exponentially without spending more on customer acquisition. Leveraging virality – as VCs call it – SaaS companies scale fast while maintaining acquisition costs that change little. Compared with SaaS, services can be a bad business, but AI startups like Megvii have reasons to target the niche. Previous key trends benefitting AI applications and secular growth potential in government IT spending and public security are now a bit clichéd. We focus on the business side here.  Services revenues weigh on gross margins as it is naturally not as scalable and has low margins. Though Megvii claims its IoT solutions involve the integration of hardware, algorithms and IoT devices, we consider this to be a service-heavy business due to the implementation model (as well as financials) falling perfectly with service business definition. Megvii city IoT solutions contributed to 73% of total revenues in 2019H1 while the gross margin sat at 59% in 2019H1. Comparing products and licenses, services have a variable personnel component that adds pressure to the margins. As of June 30, 2019, Megvii had 222 system integrators out of a total of 339 domestic customers that have contracts with the company for its City IoT solutions business.  A crowded competitive landscape. Among the top bidding-winners of the Xueliang project (Chinese official long-term security plan for cities and communities) China Telecom, China Mobile and China Unicom are dominant, followed by such large security system integrators as Hikvision, Vimicro and Tsinghua Tongfang (600100:SH), according to Chinese tech media company Tedahao (in Chinese). Some of them, along with traditional players that focus on offering hardware products, have started to enhance their software abilities. Hikvision and Dahua, two leading surveillance camera vendors in China, pivoted toward other solutions several years ago. In 2018, they announced the strategy to navigate the business – Dahua’s heart of the city (HoC) and Hikvision’s AI Cloud.  Hikvision has been catching up, for instance. Its central control product revenue has been taking up a larger proportion of the total revenue of the company, rising from 13% in 2016 to 15% in 2019H1. Dahua, a smaller one, has also been accelerating its solutions business since 2016 as well.  Megvii’s first-generation solutions roadmap was primarily driven by government use cases. It offers a full-stack IoT product that is heavily used by government agencies to enhance public security, optimize traffic management and improve urban resource planning. The government continues to be a big part of Megvii’s business, but its strategic focus has shifted toward commercial customers. The revenue mix-shift from the government to commercial poses several risks/opportunities to the business. Enterprise asks for more highly standardized products. Megvii’s gross margins are expected to rise as the proportion of personnel expenses (say, consulting, implementation, delivery) used by each new client declines. We also notice that expanding into sectors such as retail and manufacturing and building mindshare. There will be a significant advantage for a company. To solve a strategic problem and build up industry, know-how could likely yield a deep moat.  What’s the future of AI businesses and players? Part of the answer lies in the move Megvii made earlier in 2020. Megvii is trying to leverage the open-source deep learning framework MegEngine, part of Megvii’s proprietary AI platform Brain++, to allow everyone to feed data and train their AI frameworks on it. Google’s TensorFlow and Facebook’s PyTorch hold 95% of share in this market, with an array of new players joining in.  Megvii’s project Hetu is a logistics-focused platform that fits for different software systems (ERP, WMS, MES) and hardware devices (sensors, robots, AGVs) by using APIs, which has led the company to a new strategic direction. More and larger contracts with government or retail are not the ultimate goal. AI’s potential is set to change so many industries, and the best way to ride the wave is to build an operating system. The system combines data with AI approaches, like machine learning and deep learning. It keeps absorbing customer data (generated from business projects and open platforms) as well as market data (users of open-source platforms) and training these data on the system, which drives a virtuous cycle of data. As a result, that trap of data network effects mentioned before can be mitigated and a flywheel of intelligence can be set running – the model will be better, as will the product. 

Analysis
Analysis · 1
Analysis
Analysis · 1
Analysis
Analysis · 1
Analysis EO
Analysis · 2
report
Analysis EO
Jul 18, 2019 02:05 pm ·

Takeaways from Conversation With MEGVII

While there has been a lot of media hype about artificial intelligence (AI) in recent years, MEGVII is an early mover and a rare game-changer which has caught investors' attention. It recently closed its Series D funding round and raised USD 750 million from Bank of China, Industrial Commercial Bank of China, Abu Dhabi Investment and Macquarie Group. In our previous report, we noted that MEGVII has grown into China's computer vision unicorn with its globally recognized FaceID technology and it has been transforming city management, retail, logistics and many other sectors. This week, we held an exclusive interview with Li Haoran, manager of MEGVII's cloud services and spoke about MEGVII's moat and the future applications of computer vision. Key takeaways from the interview Q: From your perspective, what are MEGVII's competitive edges? How will it continue to differentiate itself going forward? A: We always have competitors in this space, but we believe the moats of our company eventually come down to two things. 1) We are an early mover in computer vision and AI commercialization. Our technology of identifying people with still postures is the first to launch in the world. 2) We are dedicated to being a vertical solution provider that solves real-world problems for our clients. When speaking to clients, we found it is not always so helpful to emphasize how cutting-edge our technology is, but rather pay more attention to their actual problems over the course of conducting their businesses. The more feedback we collect from our clients, the more advanced features we can update accordingly and the more competitive advantages we can establish compared to similar products in the market. For example, one of our clients is extending loans to consumers. When we followed up on our services, we heard that several of its customers stopped the registration process when they were required to shake their heads to confirm their identifications. The users felt that the verification process took too long even if the requirement was easy to fulfill. So we rolled out a new generation of facial recognition to capture users faces even when they keep their heads still. We aim to make the process as easy as possible while making sure the result is accurate. So far we have rolled out three versions of Face ID technology. From users manually uploading their pictures taken from different angles to users moving around their bodies in front of a camera, and finally to users standing still in front of a camera. The technology will identify their pictures automatically and verification is done within seconds. In today's rising customer acquisition costs, companies do not want to lose their prospective clients in the final step of customer identification verification. Q: You mentioned your financial client and we learned that MEGVII has more exposure in the financial industry relative to other players. Besides identification varication, what are the other applications that MEGVII provides to its financial clients? A: Our financial services clients are in banking, insurance, micro-financing and P2P lending. Our thinking is that since our company specializes in computer vision, how can we use it to help financial institutions in more situations? In speaking with client companies, we found that some of them were hoping to include computer vision in their risk management arms. Right now we are communicating with multiple clients to explore such application. Q: Can you cite examples of MEGVII's mature applications in sectors in addition to financial services? A: 70% of Android's apps have adopted MEGVII's FaceID technologies. For example, we are behind various commonly used social and dating apps. Adding identification features largely relieves users' concerns and allows them to build higher trust on the platforms. Furthermore, we have fully penetrated the domestic ride-hailing space. We help the car-hailing apps with driver verification, secure payments and secure money transfer. Uber has also approached us for similar applications. Q: Speaking about MEGVII's global expansion, we are wondering if companies in other geographic regions have more confidence in their domestic vendors? A: We have been preparing ourselves to go global for years. We have expanded to Japan, Thailand, the Middle East and Southeast Asia and our expansion has not materially been affected by the localization effects due to 1) Those countries are lagging behind in the computer vision AI technologies. 2) We do not own clients' data, so they are less concerned. Q: We think MEGVII is very strategic in choosing industry verticals. When first expanding to a new industry, how did the company overcome challenges? A: The process of zero to one is tough. When expanding to a new vertical initially, we do not have any clients. But once we obtain one, especially with those companies which have many complicated lines of businesses, we start to understand the industry and the real business needs. And then we become more effective in solving problems for clients. In the meanwhile, we have also built our brand awareness. For example, other companies who are interested in adding computer vision AI features to their companies will see China's well-known companies such as the top 4 smartphone manufacturers are on our platform. That's why companies selected Megvii even though our price is much higher than the industry average. Q: Now that you mentioned price, can you provide more colors on your pricing model? A: Most of our services are standardized. For our paid services, we use the pay-as-you-go structure so that developers pay for each individual call. For high-paying users, we also offer discounts. Even though our pricing appears to be higher other providers', it has not affected our customer retention rates. Compared to the high customer acquisition costs, the cost of facial recognition represents a relatively low portion of the overall spending of our clients. Recap of MEGVII's trajectory As early as 2011, Yin Qi was researching the topic of AI at Tsinghua University. He founded the company MEGVII with his two college friends, focusing on facial recognition. While the move was ahead of BAT and its three competitors (SenseTime, Yitu, and CloudWalk), the founding team was stuck with the next step on offering meaningful real-world applications, so they made their API public to see which companies might be interested in using this technology. Shortly after, many companies including Meitu, a photo editing app, and Jiayuan.com, a matchmaking agency, connected to the API, providing inputs to the MEGVII  founding team. A more crucial inflection point occurred in 2014, when Alibaba searched for a facial recognition provider and finally selected MEGVII to implement the "pay-with-your-face" feature in its Alipay product. Now, MEGVII is widely recognized internationally and has repeatedly defeated Google, Facebook, and Microsoft in the global AI competitions. The company changed its name to MEGVII which means mega vision, a dedicated act to transit from a technology platform to a vertical solution provider that reshapes traditional industries with AIoT (AI + IoT). MEGVII has selected three main sectors to commercialize its computer vision technology: personal IoT, city IoT and supply chain IoT.

Analysis EO
Analysis · 2
report
Analysis EO
May 13, 2019 10:08 am ·

Megvii’s Computer Vision AI Technology in 2G, 2B and 2C Segments

In today’s AI revolution, we believe the core technology comes down to Natural Language Processing (NLP), “the interpretation of human languages”, and Computer Vision, “the identification on images”. NLP is seeing broad adoption on both consumer and enterprise fronts, the best examples being Amazon’s smart speaker Alexa and Netflix’s superior recommendation systems. In China, tech giants Baidu, Tencent, and Alibaba (BAT) have been aggressively making acquisitions to roll out new innovations in the NLP field, such as Baidu’s DuerOS and Alibaba’s AliReader. Different from NLP, the Computer Vision sector in China, however, is around 70% dominated by four startups: SenseTime (商汤), Yitu (依图), Megvii (旷视) and CloudWalk (云从), according to IDC. Among more than 5,000 AI companies in China, almost 90% of them are simply using the “AI concept” without creating value-added solutions for clients. Some startups did gain leading technologies in image and video recognition after years of research and development, but selling SDK alone doesn’t bolster their monetization capabilities; nor does it guarantee that they will beat the competition. SenseTime, Yitu, Megvvi and CloudWalk, on the contrary, have become unicorns with billion-dollar valuations. This has resulted not only from their technological advantages, but is driven by the end-to-end vertical solutions they can provide to businesses and governments. Megvii is an early mover with globally recognized technology As early as 2011, YIN Qi (印奇) was researching the topic of AI at Tsinghua University. He founded the company Face++ with his two college friends, focusing on facial recognition. While the move was ahead of BAT and its three competitors (SenseTime, Yitu, and CloudWalk), the founding team was stuck with how to offer meaningful real-world applications, so they made their API public. Shortly after, many companies including Meitu, a photo editing app, and Jiayuan.com, a matchmaking agency, connected to the API, giving ideas to the Face++ founding team. A more crucial inflection point occurred in 2014, when Alibaba searched for a facial recognition provider and finally selected Face++ to implement the “pay-with-your-face” feature in its Alipay product. Now, Face++ is widely recognized internationally and has repeatedly defeated Google, Facebook, and Microsoft in the global AI competition. It has ranked No. 1 on 17 technology evaluations worldwide and has been featured in MIT Technology Review three times. Transforming traditional industries with Computer Vision Today, Computer Vision technologies have been applied to multiple industry verticals: mobility (autonomous driving), healthcare (medical scans reading and disease detection), retail (facial recognition to implement personalized advertising), finance (anti-fraud, identity verification), logistics (production line monitoring), and public security (anti-terrorism, video surveillance), and so on. Face++ will focus on three major sectors in the next three to five years: smart city, supply chain, and lifestyle. We believe the company is very strategic in choosing industry verticals that can be empowered by Computer Vision technology. The company changed its name to “Megvii” meaning mega vision, which symbolized a commitment to transit from a technology platform to a vertical solution provider that reshapes traditional industries with AIoT (AI + Internet of Things). Servicing governments - “Smart city AIoT” Evolution: Before the concept of “Smart City” kicked in, Megvii tapped the public security industry in 2014, as facial recognition was likely the “must-have” technology for the sector: security apparatuses often need to verify people’s identities without their cooperation. Traditional surveillance systems are cumbersome to install and relay low-definition imagery at a slow speed. Therefore, since the four Computer Vision unicorns started offering high-resolution cloud-based video surveillance systems with intelligent analytical capabilities, an increasing number of police departments have become their clients. According to the company, Megvii secured contracts with police in 32 cities in China and has assisted in over 5,000 arrests of wanted suspects in two years. Financially, the smart surveillance business contributed around 70% of total revenues and grew nearly 50 times in 2018. Public security industry: The Chinese government has plans to build a closed-circuit television (CCTV) surveillance network for public security purpose, with 170 million security cameras in use and another 400 million being installed. While the controversy on potential privacy invasion is not within the purview of this report, we believe the government decisions make China the largest market in video surveillance.The US market, worth an estimated USD 2.9 billion and growing at 0.7% a year, pales in comparison to the market size in China, which reached USD 6.4 billion in 2016 with an annual growth rate of 12.4%, as reported by HIS Markit.Since traditional surveillance producers still take over 90% of the market share, we anticipate boundless growth opportunity for Megvii. Current challenges: While the system recognizes real-time unlawful activities, it cannot act in a timely fashion to stop such activities, making the AI technology less “intelligent” than expected. Moreover, since there are only two engineers with more than 10 years of experience in the security field at Megvii, the company is still in dire need of industry expertise at this point. Going beyond defense: In addition to aiding criminal investigations, Megvii eyes the opportunity to build “City Brain”, acquiring larger data sets and empowering entire cities with the Computer Vision technology, specifically regarding urban planning, traffic, and large-scale events. The company produces hardware including the smart portrait capture machine MegEye-C3S, the face recognition terminal MegBox B1R, portable devices, and card readers, to capture not only individuals, but also roads, automobiles, buildings, and so on. Its software then extracts actionable information from the images/videos and delivers analytical results. Currently, the company is solving issues of randomly parked bicycles in Luohu District, Shenzhen and providing AI solutions to the World Internet Conference held in Wuzhen annually. Global expansion is also on the cards: In 2018, the company sold its Smart City solutions to over 20 countries on four continents and undertook multiple projects in Middle Eastern countries. Servicing businesses - “Supply chain AIoT” Compared to the smart surveillance business, the supply chain AioT is still at an early development phase in which Megvii is testing and creating market demands. In 2018, the company launched its Supply Chain solutions and claimed it was the primary focus. Linking logistics with Computer Vision: Though logistics seems irrelevant to Computer Vision, Megvii hopes to transform the sector with its Robotic Network Operating System, Hetu, to automate most logistics work. For example, it has helped Tmall Supermarket’s warehouses boost efficiency by almost 40% and shorten the duration from receiving orders to making ready-to-ship goods to below one hour. Last year, it acquired Ares Robot (艾瑞思机器人), a producer of robots that replace humans in picking and sorting packages in factories as well as offers cloud-based warehouse management system. Moving downstream to retail: With solutions to optimize the supply chains, it is natural for Megvii to move a little downstream to the retail sector. Currently, KFC, Tao Café, Raffles City (来福士), EATOWN (怡食) and other retail businesses are using Megvii’s retail solutions to digitalize their offline stores and transform themselves to New Retail. When the consumer walks into the store, Megvii’s smart terminal will identify his or her age, gender, and other dimensions, whereupon the shelf sensors will track their purchasing behaviors. Combined with existing ERP/CRM, Megvii’s retail solutions allow store owners to understand what goods should be purchased and when to put them on the shelves and help them design precision marketing. Such solutions have already made impacts on Good Neighbor, a well-known supermarket chain in China, by increasing its inventory turnover and weekly repurchase rate by around 100% each. Servicing consumers - “Lifestyle AIoT” Megvii’s facial technology is already ubiquitous in consumers’ lives, from social media to mobile payments. For instance, it offers face unlocking, image beautifying, location optimization, and other services to Xiaomi, Oppo and Vivo as well as secure payment and remote account opening services to CITIC Bank and Ant Financial. Valued at USD 4 billion ahead of its upcoming IPO Last week, Megvii announced that it had raised USD 750 million from its Series D funding round, giving the company a valuation of USD 4 billion. Likely due to the sensitive nature of its business, the majority of investors are state-owned organizations like Bank of China Group Investment, ICBC Asset Management Co., and a subsidiary of Abu Dhabi Investment Authority. Other investors included Macquarie Group and Alibaba. The company is planning to file for an IPO on the Hong Kong Stock Exchange later this year. Looking forward, we are optimistic about the company’s potential to reshape multiple industries in China, given its advanced technologies, clear commercialization path and top-tier management team with vision and execution ability.

Analysis EO
Analysis · 2
Analysis EO
Feb 2, 2019 11:35 pm ·

Megvii Launches The Largest Checkout Dataset

Megvii (旷视科技), previously known as Face++, claims its Nanjing based research center has developed the largest scale of retail product checkout dataset name as PRC for retail ACO  (automatic check-out) in terms of the number of images and categories. This dataset aims at increasing efficiency in the retail industry, which is relatively labor intensive. Megvii suggested that costs occurred during checkout constitute a large proportion of the total operating costs of retail outlets, and they see it is a trend to reduce the checkout costs by integrating computer vision into the retail industry to achieve ACO. ACO is essentially an automatic solution that will automatically generate a shopping list from the images of products. The supposed benefits from applying an ideal ACO system are shortening the checkout time and lowering staff costs as it only requires computer vision technology to identify total products purchased and sum up the total payment. Although ACO might sound like simple, it is not well studied in the computer vision community currently and faces the following challenges at the current stage listed as a result of the lack of high-quality dataset : Lack of large scale of high-quality products images The number of products could be large, or photos were taken in an environment different from the deployment scenario. Therefore, there is insufficient appropriate training photos pre-product. Fine-grained nature of the product categories Sometimes different products could share a great level of similarities, and products are subject to constant package changing. Thus, it will add some difficulties for computer vision to identify correct items. Difficulties in collecting training images Since the number of products is huge, it might be impractical to collect all sufficient images of each individual product. Meanwhile, the training images reflecting the realistic checkout scenarios is also insufficient   In order to improve the current ACO technology, Megvii tried to mimics the real-world ACO scenarios. The first thing, Megvii tried to solve was increasing the number of images and product categories. The new data set proposed by Megvii contains 200 product categories and 83,739 images. To increase image recognition accuracy single-product images were taken in a controlled environment and multi-product checkout images taken at the checkout counter with various annotations are provided. Secondly, they use exemplar images and checkout images to increase accuracy. As in exemplar images are basically for capturing multi-view appearances for every single product, while checkout images are mainly for gathering realistic checkout scenarios where each image includes multiple products. Thirdly, in order to mimic the real checkout scenarios, products are randomly chosen, combined, and freely placed on the checkout background with random orientations. Occlusions and complex clutter are widely applied in Megvii’s RPC dataset. Fourthly, RPC’s in a hierarchy structure meaning their 200 SKUs can be categorized as 17 meta-categories which cover diverse appearance such as bottle-like, box-like, canister-like. The SKUs under each meta-category tend to be fine-grained. Fifthly, according to the number of products and product instances, images are split into three categories of easy mode, medium mode, and hard mode. For each clutter level, there will be a clutter level annotation attached which allows in-depth inspection of the model capacities. Sixthly, for each individual RPC image, there are three types of annotations from weak, medium to strong. These three types of annotations are divided into shopping list, point-level, and bounding boxes. Shopping list is said to be the weakest level of annotation and the easiest to obtain in practice. And it records the SKU category and count of each product instance in the checkout image. Point-level annotation, which provides the central position and the SKU category of each product in the checkout image. Bounding boxes, the most labor-intensive annotation, provide bounding box and SKU category for each product. For the evaluation, it adapts 4 protocols which are Checkout Accuracy (cAcc), Average Counting Distance (ACD), Mean Category Counting Distance (mCCD), Mean Category Intersection of Union (mCIoU). CAcc is the most important metric for the ACO task according to Megvii team, which is used to measures the pass rate of an ACO system and thus reflects the practicality of the system. Meanwhile, ACD only considers the average number of counting errors, mCCD measures the average ratio of counting errors, and mCIoU measures the compatibility between the predicted shopping list and ground truth. Moreover, Megvii came up with four baselines, which are Single, Syn, Render and Syn+Render to benchmark the dataset. It is to be noted that the data set came up by Megvii is more advanced than current datasets in this field, while there is substantial room to improve ACO performance.

Analysis EO
Analysis · 2
Analysis EO
Jan 24, 2019 08:19 pm ·

From Face++ to Megvii, The Company Needs A New Story

Megvii, used to be known as Face++, announced in its annual strategic conference last week that the company would invest CNY 2 billion (USD 296 million) into developing artificial intelligence based supply chain solutions provider through its new flagship product Hetu, EqualOcean reported on Jan 17. Before the announcement, the company’s brand name was Face++, which is a face recognition product used by popular consumer apps including Alipay (支付宝) and Meitu (美图), the dominating mobile payment app in China and the popular beauty enhancing photo app. The company chose to abandon this widely-known brand name and use the current name Megvii, trying to express the determination of its strategic transformation from a face recognition technology provider into an AI-based supply chain solutions provider. The new flagship product Hetu, named from China’s fairy story of creation, is a robotics operating system enabled with AIoT (AI + Internet of Things), having the ability to provide whole sets of software and operating system solutions for manufactory, logistics, and retail industries. In another world, Megvii is no longer a simple AI technology company, but now a solution provider for the supply chain industries of the world’s factory, China. As mentioned above, Megvii’s previous flagship product Face++ was popular among consuming application developers, as it provides accurate and convenient face recognition SDK (Software Development Kit) based on the company’s advanced computer vision technology. But now, the company has realized that the market scale of consuming application SDK development is way smaller that logistics and supply chain intelligent operation system, while the two aspects use the same technology, computer vision. In October 2018, Ma Huangteng (马化腾, Pony Ma), the co-founder and chairman of the board of Chinese tech giant Tencent, asked a question on China’s version of Quora, Zhihu, that how the industrial Internet would impact on the consumer Internet in the next 10 years. The question that Pony Ma raised caused a stir among the tech community in China and received more than 3000 answers. People realized that as the most valued consumer Internet company in China, Tencent is now pursuing new opportunities in the industrial Internet. By transforming the company’s industry direction, Megvii is in-fact telling a new story, that it has joined the trend of industry along with tech giants such as Tencent and Alibaba. The market scale of China’s supply chain management service industry will reach CNY 378 trillion (USD 56 trillion) by the year of 2023, according to a market report by Qianzhan Industrial Research. Even though Megvii has entered this enormous market, it’s still unsure if it is able to assume a stake from it. But its transformation might be a good story to the potential investors of the company. The latest fundraising of Megvii happened in October 2017, when investors led by China State-Owned Assets Venture Investment Fund and Alibaba injected the company USD 460 million and valued it at USD 2.3 billion.  In July 2018, it was rumored that Alibaba was ready to finance Megvii’s Series D funding round, valuing the company at USD 3 billion. But it hasn’t been verified by both sides of the matter. The timeline indicates that the company hasn’t closed any funding round for 16 months, a relatively long period compared to top AI startups’ average financing cycle. While during this period, Megvii’s main rivals, SenseTime, Yitu, and CloudWalk have all closed at least one large investment round. The four startups were once regarded as the “Four Computer Vision Tigers,” and all of them except Megvii have completed the transformation of their role from computer vision technology platforms to commercial delivery platforms. SenseTime released its face recognition algorithm platform SenseFace 3.0 and city large-scale video analysis platform SenseFoundry in the middle of 2018, signifying that the company has turned its direction into city security. The two platforms help the police force to analyze images and videos captured by thousands of cameras in the cities. The company raised USD 1 billion from Softbank China in September 2018, valuing it at USD 6 billion, making it the world’s most valued AI startup. In August 2017, Yitu, an image recognition company, announced for the first time that the company was entering the healthcare field, by releasing its AI + Healthcare One-Stop Research Platform, focusing on the analysis of medical images by using computer vision and deep learning technologies. The company raised USD 300 million in its Series C round of funding in the middle of 2018, valuing it at USD 2.4 billion. CloudWalk was founded as a face recognition SDK platform in 2015, the company turned its main business into financial risk management after it initially worked with the Bank of China in May 2017. The company now classifies it as a financial technology company which provides intelligent financial service solutions by using artificial intelligence technologies. CloudWalk closed its Series B+ round of funding which valued it at USD 3 billion. From this point of view, it’s not hard to see that only by transforming ones’ direction of development from technology providers into commercial delivery platforms will these computer vision startups be valued at a higher level. Megvii seems to have realized the situation, and it’s not too late.

Analysis EO
Analysis · 2
Analysis EO
Jan 21, 2019 08:00 pm ·

Megvii's Principal Scientist Dr SUN Joined Xi’an Jiaotong University

Jan 21, 2019 /EqualOcean/ - According to Xi'an Jiaotong University's Official Weibo Account, the university announced the establishment of the Artificial Intelligence College. Dr SUN Jian (孙剑), the principal scientist of Megvii (旷视科技) and also previously the part-time professor of Xi'an Jiaotong University, becomes the dean of the new college. This shows the technical capability of Megvii is advanced and reliable. Also, this is the first step of the cooperation of Megvii and Xi'an Jiaotong University, which might bring benefits for both of the university and Megvii in the future. Located in the ancient capital Xi'an, Xi'an Jiaotong University is one of the top universities in China. The university is already famous for its Artificial Intelligence And Robots Institute which is founded in 1986, and Dr SUN Jian is an alumnus of the institute as well. The Artificial Intelligence And Robots Institute is one of the earliest and leading institute in China that is investigating on artificial intelligence. However, the institute only provides Master and Doctor degrees, the new Artificial Intelligence College will help fill the vacancy on the education of the undergraduates. Actually, Xi’an Jiaotong University already planned the undergraduate education of artificial intelligence in 2017. April 2018, the first artificial intelligence experimental class is established, some top students from Electronics and Information Engineering are chosen to be the first batch of the experimental class. Roman is not built in one day, now the Artificial Intelligence College accepts students nationwide. Then, why do Xi’an Jiaotong University establish the Artificial Intelligence College at this point? Actually, the establishment does not happen singly, China’s universities are establishing the artificial intelligence college in a rush. As we can see from the figure The Chinese Univerisity's Artificial Intelligence College Founded Year below, only 1 artificial intelligence college is founded in 2016, 7 colleges are founded in 2017, and 27 artificial intelligence colleges are founded in 2018 which is almost three times more than the establishment in 2017. The fast increment is actually the consequence from government and university. The Politics April 2018, the Ministry of Education of the People's Republic of China (MOE) has released the Action Plan on Promoting Artificial Intelligence in Universities, which shows its plan to establish 50 artificial intelligence colleges or institutions by 2020. Under the instruction of MOE, the artificial intelligence college has rapidly set up in 2018. Xi’an Jiaotong University is one of the leading universities in China, which follows the politics and establish the artificial intelligence college. The Universities In the classical model, the universities provide the education to its students under the instruction of MOE. However, this might exists the gap between what the universities teach and what the is really practical for the enterprises. This wave of the establishment of artificial intelligence colleges breaks the old rule. There are three universities cooperate with Tencent Cloud (腾讯云) for the establishment of the colleges: Shandong University of Science and Technology (山东科技大学), Liaoning University of Engineering and Technology (辽宁工程技术大学) and Liaocheng University (聊城大学). Chongqing University of Posts and Telecommunications cooperates with Iflytek (科大讯飞). Also, in this news that Xi'an Jiaotong University cooperates with Megvii. This shows the trend that the universities start to cooperate with enterprises, and the university can directly teach the students related technology need in the society and enterprises. The enterprises will give the instruction on the university courses. Last but not least, the cooperation of both parties: Megvii and Xi'an Jiaotong University could lead to a win-win result. Megvii have the chance to educate the students in Xi'an Jiaotong University and this will give feedback to improve the technology in Megvii as well. On the Xi'an Jiaotong University’s side, the university could improve its education quality and the employment rate of its graduates. The students in the new college will also become valuable talents that will be needed by society.

Analysis EO
Analysis · 2
Analysis EO
Jan 17, 2019 09:44 pm ·

Megvii Released Its Operating System Hetu And Upgrade Its Strategic System

Jan 17, 2019 /EqualOcean/ - The Megvii (旷视科技) Robot Strategic Conference was held in Beijing on January 16, 2019. Megvii released its brand new flagship product: Hetu (河图). Hetu is a robotics operating system enabled with AIoT (AI + Internet of Things). That is to say, attached with Megvii’s artificial intelligence technology, Hetu is able to connect with IoT and provide a whole set of solution on planning, simulation, application and operating, The name “Hetu” came from the fairy story of ancient China, the god and ancestor of Chinese Fu Hsi (伏羲) inspected the sky and nature, he drew the map named Hetu after inspection. Later on, Eight Trigram and the Book of Changes developed based on Hetu, and the Chinese civilization developed gradually. The name “Hetu” represents Megvii’s expectation and ambition on this flagship product, In this Megvii Robot Strategic Conference, Megvii also announced the abandon of its old brand and logo “Face++”. This shows the strategic meaning for the company Megvill. Actually, the logo “Face++” is already a very popular brand in China, as Megvii provides the facial recognition services through the brand “Face++”. Facial recognition services are very popular in China, as China has the largest amount of population and the females are fond of taking selfies in the Chinese modern pop culture. Hence, when people tries to take selfies, the customers are used to using Meitu (美图) to help with their selfie beatifying, As Meitu is utilized with the AI and facial recognition technology provided by “Face++”, the brand hence is famous in China. Later on, when Alipay decides to do the Smile to Pay function, Face++ is selected as its co-partners for its leading technology in facial recognition. Since then, Face++ became the brand that represents the leading technology in facial recognition, and people would like to call Megvii as Face++ for the popularity and widely acceptance. In the Robot Strategic Conference, Megvii announced officially that it will stop using the name “Face++” and even replace the old classic logo “Face++” into the new logo “Megvii”. This action shows the determination of Megvii to abandon the classic impression of the artificial intelligence company with leading technology only in facial recognition. The abandon of the classic logo may lose the related influent power, and this big move shows the strategic upgrade of Megvii. Founded in 2011, benefiting from the mobile internet, Megvii develops rapidly into an AI unicorn by providing the facial recognition technology to companies like Meitu and Alipay. Megvii is a very smart company and it always target the technology that could apply and bring the revenue successfully, with the financial accumulation from facial recognition, Megvii further developed its technology to computer vision. Different from facial recognition, computer vision usually provides technology 2B instead of 2C. That is to say, the technology facial recognition is to identify and analyze people’s face through artificial intelligence, computer vision is totally switch to the industry side and it is the eyes for the machine and robots. For instance, when the workers make the component in the factory, there is no doubt that the allowance should be permitted on the component due to the variant operations during the process, By applying the computer vision, the robots or the machines has the ability to control the allowance and this improvement will bring the factory two main benefits. The component quality can be highly improved and the raw material can be saved which provides more profitability. April 8, 2018, Megvii acquired the roboticis company Ares Robot (艾瑞思机器人), which shows the technology upgrading from facial recognition to computer vision. Ares Robot is famous for its intelligence warehouse solution (iWS), the cloud based warehouse management system iRMS, intelligent warehouse robots iWR500 and other related technology. By successfully acquire Ares Robot and combine the algorithm of mature facial recognition technology with the warehouse technology, Megvii successfully upgrades its artificial intelligence technology to computer vision. Hetu is the operating system based on the technology of AIoT. In the application of Hetu, digital twin is the required AI technology. Then what is digital twin? It is a technology that could provide a digital world run through data and algorithm, and compares with the real situation in physical world. The application of digital twin can provides several beneficial results. For instance, if the factory wants to build a new production line. Supported with digital twin technology, people can build the digital one first, so the factory can encounter the issues and problem in the digital world first, this could decrease the building time of the production line in the physical world. Besides of this, in real world, people could make mistakes during the working process, and the risks are mainly random. By comparing the difference between the digital world and the physical world, digital twin can identify the risks very fast and precise. IoT is easy to understand that it is mainly related with the connection to the robots and the related control. To be more precise in the technology, when applied to industry side, AIoT can be separated into three parts: the identification, the control and the optimization. Megvii has the leading technology in the identification and optimization part with its successful algorithm in facial recognition. Whilst the acquisition of Ares Robot improve its ability in control section. Megvii is very strategic not only in the technology part, but also in choosing the industry. By choosing logistics as its primary industry, Hetu can relief the hard-working of the workers that works repeatedly in the factory and logistics warehouse. Also, logistics is actually a widely distributed industry in China that even companies like Jingdong is highly depends on logistics as well. Megvii claims that Hetu will find the best co-partners in the industry to build the operating system, and the system will open to all applications. This is a very brave and smart try to apply AIoT in the logistics industry, however, it is always difficult for the internet companies to apply the technology in the industry. So, it is very look forward for the author to see, the strategic upgrade is a chance or a risk for Megvii.