Cable for apple
The cable for apple also call iphone cable,USB C apple,USB C to lightning, lightning cable.This USB Cable connects your iPhone, iPad, Mac, AirPods or iPod with Lightning Connector to your computer's USB port for syncing and charging. You can connect to the Apple USB power adapter for convenient charging from a wall outlet. 1 meter in length.Wisesir provide a variety of cables for Apple accessories, high performance chips, support fast charging.
Cable for apple,Usb C to Usb C,Data Link Cable,Usb Data Cable Price Pogo Technology International Ltd , https://www.wisesir.net
What is the background of the Chinese company who won the Microsoft Face Recognition World Cup Championship?
(Original Title: What's the Background Story of the Chinese Company That Won Microsoft's Face Recognition World Championship?)
Xia Bi Oblique
In June 2016, Microsoft publicly released a large-scale real-world facial image dataset, MS-Celeb-1M, which contained approximately 10 million (10M) facial images representing around 100,000 celebrities. This move encouraged researchers to leverage these datasets to develop and test face recognition technology.
At the same time, Microsoft announced the MS-Celeb-1M Million Face Recognition Challenge. Participants were tasked with developing image recognition systems using the challenge-provided datasets as training data and identifying one million celebrities from facial images.
In essence, participants were challenged to identify celebrities from photos. The use case is highly relatable to everyday life and can easily integrate into business environments, such as image tagging and news video analysis, public opinion analysis, and more.
Even from Microsoft's perspective, the MS-Celeb-1M challenge was significant. The dataset effectively bridged the gap between industry and academia, and through the design of targeted evaluation metrics, the competition helped face recognition models transition into practical applications.
This year, a Chinese startup took first place in the list of millions of people identification challenges. Their name is Orion Star.
Limited Data Competition, Algorithmic Prowess, and Product Logic
With the support of the ICCV 2017 platform, a top-tier computer vision conference in the AI world, Microsoft launched Hard Set and Random Set competitions, as well as a new, more challenging small sample learning (Low-Shot Learning) event. This was considered a "high-level, most difficult" technical event.
In this competition, the unrestricted category wasn't just about simple algorithms. To some extent, it was about how much data the trained model could handle. The more data, the better the unrestricted category became a competition of algorithms and data; while the restricted category was purely about algorithmic prowess, making it even more challenging. As PingWest understands, "Accuracy solely depends on the algorithm when data is fixed. Without external data, achieving the highest accuracy becomes incredibly difficult." Therefore, the organizing committee deemed it extremely difficult and listed a separate sub-project with a restricted dataset.
Thus, in the projects involving Orion Star, external data couldn't be used. Restricted categories allowed only the use of a limited dataset and competed on the team's algorithmic capabilities.
In short, it was about achieving the best results with minimal resources. The difficulty lay in both the technical algorithms and the overall product design thinking of the team.
According to Orion's introduction, the challenge with limited datasets was that the competition data was very noisy and couldn't be directly used for training. However, the entire training dataset was vast. Models required for large-scale face model training were resource-intensive, making it hard to perform efficiently and effectively.
"Training data is dirty. By dirty, we mean there are many different people in a category. Sometimes, it's unclear who the correct person is within a category. Training this type of data directly is challenging," PingWest learned.
After research, the Orion Star team designed a robust denoising algorithm that could effectively clean up various types of noisy data. They also chose a triplet model suitable for large-scale training and innovatively accelerated triplet network training under limited resources, significantly improving performance.
The Orion Star team eventually scored 0.75/0.606 (random set/hard set). According to the competition organizers' words, 0.75 is close to full marks and a remarkable achievement. The 0.606 score on the hard set surpassed last year's best score (0.534).
One question arises: how did a company just established in 2016 manage to achieve such results in this event? What's the background story of this company?
Behind Orion Star lie the two core technologies of image and speech recognition.
The challenges faced by AI technology are now evident - academic talent continues to publish papers, but it's not always possible to apply these technologies in large companies and bring them to market quickly; what's currently lacking is industry talent. These individuals face the challenge of translating papers and academic knowledge into practical product applications. Actually, opening the application market for AI technology pays more attention to the latter.
Orion Star plays this latter role, allowing technology to take root and become closer to consumers. It already has several floor products.
Searching for company information - Orion Star's full name is Beijing Orion Star Technology Co., Ltd., founded in September 2016, it's an AI company under Cheetah Mobile. At its inception, Orion Star gathered global AI talent. The startup team included tech experts and product geeks from top-tier U.S. technology companies, including Silicon Valley, Japan, Taiwan, mainland China, Beijing, and Shenzhen.
In its early days, Orion Star focused on researching and exploring face recognition. They achieved top-three results in this year's LFW face recognition evaluation. Unlike teams that solely seek precision, Orion Star uses small networks to achieve as high precision as possible. For product applications, their strategy is to quickly apply to products.
It is understood that - on this face recognition technology, Orion Star will be used in actual scenarios and ultimately will be implemented in access control, robotics, mobile apps, and other specific products.
Its face recognition technology has been applied to Live.me, a live streaming product from Cheetah Mobile. The main function of Live.me is to identify inappropriate content. The live streaming platform generates over 200,000 hours of live content daily, using image recognition technology for 24/7 real-time monitoring, greatly improving work efficiency and outcomes.
In terms of voice recognition, in June this year, Orion Star jointly launched the Xiaoya AI speaker for users with Himalaya. This is the first product on the Orion Star AI ecosystem. The core behind the Xiaoya AI speaker is the far-field voice interaction technology developed by Orion's full suite. This system creates a perfect user experience in areas like responding to calls, high-quality on-demand experiences, and massive content.
Just over a month later, Xiaomi also released Xiaomi's first AI speaker, "Xiaoai Teacher." This is the first collaboration between the AI company Orion Star and Xiaomi from Cheetah Mobile. This product provides human voice response technology like speech synthesis.
In short, when Xiao Xiao classmates were awakened, the answering content like “Hey†and “I’m here†were provided by Orion Star, while the speech synthesis technology was the speech model when processing speech feedback. In tone, it made her sound less robotic and more like a natural person. Through the TTS speech synthesis mode, you can also let your favorite star call you up every day with its tone and tone.
The combination of rapid technology and product application is the core appeal of this product-oriented AI company.
Fu Sheng’s AI Thinking and Cheetah’s Transformation
According to Fu Sheng, CEO of Cheetah, Internet+ was the last wave of cognitive dividends, and the next is AI. The AI+ era has arrived.
The Internet industry has changed the news+Internet that we traditionally recognized, evolving into a larger Internet world. AI artificial intelligence technology will also reconstruct the entire industry and reshape our entire way of thinking.
"AI is just like the Internet. Is the Internet fundamentally a technological breakthrough? It is because after this agreement was created and websites were built, many product forms changed. What you see today, for example, is payment. We also call it Internet+, all industries are on the Internet, and today all the dividends of the Internet come from the combination of traditional industries and applications. We really think about the first wave of the Internet as actually an Internet+, combining news with the Internet, then combining content with the Internet, entertainment, and the Internet, it is a basic technological approach."
So AI I think is also such a thing. It actually can change the industry rules of many industries. It is hard to become an independent industry in itself and it must be combined with applications. So the arrival of the AI+ era."
Artificial intelligence attributes all physical events to one point—data. Then, this data is used in a neural network to understand and comprehend, reaching a level that all algorithms in the past could not reach.
Fu Sheng believes that the opportunity for deep learning lies in the combination of application and not technical output. The core of deep learning is data-driven. Although there are model adjustments and have their own advantages, others have more data adjustments to quickly level the advantage. It is difficult to really imagine that a company can succeed by providing technical output.
Under Fu Sheng's thinking, Cheetah has begun to lay out content products.
The earliest positioning of Cheetah Mobile was security and tools. After successfully commercializing the tool products, Cheetah Mobile began upgrading its track.
In addition to the live product Live.me in the US market and News Republic, the news service operator, Cheetah also invested in the Musical.ly short video team.
According to the financial report for the first quarter of 2017, the contribution of content-based applications to Cheetah’s total revenue increased from 11.3% in the previous quarter to 18.8%, a 55.2% increase from the previous quarter. This is thanks to the outstanding performance of its content-type products, such as Live.me and News Republic, in overseas markets. Both products use artificial intelligence technology.
This Orion Star company should be the new driving force accelerating Cheetah's pace.