Create AI exclusive fashion consultant mother never worry I will not dress
(Original title: After accumulating Asia's largest female figure database with virtual fitting clothes, good clothes want to use AI as everyone's exclusive fashion consultant)
Once, virtual fitting and humanoid robots are human dreams that have always been difficult to achieve. However, this also means that once they are realized, they will greatly change our lives and at the same time bring huge returns to the innovators.
This "temptation" also prompted Chai Jinxiang to start business from the academic world. In 2000, he worked at Microsoft Research Asia, where he published the first SIGGRAPH (Top Conference on Computer Graphics) papers of the Institute. Later, he went to the United States to study for a computer science degree from Carnegie Mellon University, and at TAMU. A tenured professor. Now, he is the co-founder and CTO of the virtual clothing company.
Good Maid was established in August 2013. Since its establishment, it has cooperated with a number of Tmall TOP 50 women's brands to provide them with online virtual fitting services. In addition, the original measurement-free measurement technology has been established. Accumulated 4.8 million real user-entered figure data; at CES Asia in June, Goodclothing also announced that it will cooperate with C&A to launch an artificial intelligence (AI) offline experience in the second half of this year. Stores, building omni-channel solutions.
Since its inception, Haoyiyi has obtained angel rounds and A-round financing from Lenovo Star and Zhiping Capital, and in 2015 it received a $15 million round B round of capital financing from Broadband Capital. The reporter interviewed Chai Jinxiang and learned about the different virtual dressing routes.
Using big data to solve the problem of virtual fitting
Chai Jinxiang’s main research areas are human animation and motion capture. The former studies game, movie and human-related 3D animation. The latter field studies real-time capture of human movements. In recent years, the hot camera uses the Kinect camera to capture human posture. It is one of his directions.
These research directions are closely related to virtual fitting. All the time, researchers and commercial companies have tried virtual fitting. Historically, there are two main technical routes for virtual fitting:
First, use traditional animation methods to do this; this is also a common method used in games and cartoons, a piece of clothing, first manually modeled by the artist, to make the shape, color, texture of the clothing, and, due to this "Clothing" is worn on different people and simulations are also performed.
This is the most classic practice of virtual fitting, but there are also many drawbacks: First, the effect is not real enough; second, and most important, the cost of this approach is very high, manual modeling requires a high level of art, and only rendering It may take a week for a piece of clothing. Seasonal play of clothing is fast, and this practice can't be done at all.
Second, with the AR method, the pictures of the clothes are “sticked†onto the pictures of the human body; the pictures can guarantee that the clothes are more realistic, and the cost of this practice is lower, but the “texture†also means the virtual fitting. The effect is greatly reduced.
Is there a new method that can make up for the deficiencies of the two traditional methods?
The practice of buying clothing is based on computer graphics and computer vision, which is already a category of artificial intelligence.
Virtual fitting involves the three-dimensional modeling of clothing and human body: clothing, good clothing, computer graphics and computer vision methods, first take pictures of the clothes from various angles, and then model it through algorithms. Chai Jinxiang told Lei Feng network to create new wisdom, at present, the shooting of clothes is a standardized process, a set of clothes can be taken in a few minutes.
In order to ensure the fitting effect, it is equally important to perform 3D modeling on the human body. Now many startups use the Kinect depth camera to capture the posture of the human body. However, a good solution is to use a different type of clothing: In combination with the model in the human body database, plus the user's active input adjustment, automatically generate any three-dimensional model of any body, without any aids such as ruler, camera, depth camera, etc. equipment.
Good Buying will call this program a “free measurement,†and Chai Jinxiang tells Lei Feng that it has created new wisdom. In order to achieve better results and user acceptance, good clothing has been tried in three phases. The first phase is good to buy. The clothes launched an app called "size camera". Users can use the mobile phone to take photos in three different positions. The system can model the three-dimensional data of the human body. However, because the cost of download and operation is relatively high, the app does not end up. Can be accepted by the user, "100 users may only have one to download and use"; then, good clothing tried the way of big data prediction, the user can input height, weight and measurements, the system has built a model in the database to build However, in the actual process, many female users did not know their own measurements. Finally, it was a good idea to explore the current model. Users only need to input height, weight, and cups of underwear. More than a dozen body tags are selected for modeling, such as "There is no belly," "Big belly is not big," "Tough thighs are not thick," and so on, dozens of seconds In time, the system can simulate the human body.
In addition, good clothing also fully takes into account the C-end user's habits, allowing users to upload their own front face, the system can automatically identify, locate and extract facial features, through the three-dimensional reconstruction of the synthesis technology to restore the user's true facial features to virtual 3D people Face model. Users can also customize their favorite hair, hair color, face, skin color.
Whether it is relying on image modeling or measurement-free measurement, tuning algorithms are very important tasks. Chai Jinxiang told Lei Feng network to create new wisdom, from the establishment of good clothes to the introduction of current products, it took two or three years. In March 2017, Tmall opened a new fashion event, and the virtual fitting provided by the good clothing has become the biggest highlight of the women's meeting place. "Cooperation with Tmall also means that our effect has been recognized."
Exclusive Fashion Consultant: AI recommends suitable clothing according to body size and appearance
Now, besides the virtual fitting, the good clothes also decided to do something more artificially intelligent: to do everyone's exclusive fashion consultants, that is, through the user's body and appearance, actively recommend suitable clothing.
Like the virtual fitting, the outfit recommendation involves both clothing and people. Chai Jinxiang expressed that the former uses a deep learning algorithm to acquire clothes. After a large amount of data has been manually annotated, the system can now automatically model the same. Good clothes models are marked with various labels, such as fabrics, styles, styles, details, etc. This practice can be carried out smoothly, thanks to the standardized process of clothing modeling for clothes. Structural data after modeling can be used directly to optimize deep learning algorithms.
The human factor is to design the figure and appearance. Although it is a pure researcher, Chai Jinxiang has already had a set of experiences on clothing. “A person with a shorter lower body and a thicker thigh, for example, is more suitable for wearing an A-line skirt. Its waist line is relatively high, it will appear that the legs are longer. Wearing is actually to avoid weaknesses."
But for machine algorithms, there is still a lot of difficulty in recommending clothes according to their size and appearance. The biggest difficulty is that these “knowledge†or experiences do not have a systematic, documented system.
The advantage of good clothing is that it has a lot of figure and outfit data. In the Tmall app, good clothing and a large number of TOP 50 women's brands have reached cooperation to provide on-line fitting services. Chai Jinxiang said that at present, 4.8 million users have used this function to enter body data and carry on the experience. This is a great help in recommending clothes according to your size.
People's faces and appearances also have particular emphasis on outfits. However, this is a more difficult system to solve. The good clothing plan is to cooperate with the top models in the country to transform their experience into a line. Effective algorithm.
Compared with virtual fitting, fashion consultants are still not mature enough. Chai Jinxiang scored 60 points for him. Although he still has a long way to go, he can imagine that AI consultants have unparalleled advantages compared to real fashion consultants. "There can be no 10,000 clothing recommendations in the mind of fashion consultants. Our system is more powerful."
In the future market strategy, good clothing will also shift from focusing on virtual fitting to focusing on omni-channel solutions. On the one hand, it will provide virtual fitting services for online stores; on the other hand, it will cooperate with brand partners to create offline experiences. Shop, fitting mirror will be a very good education market product, it can also cooperate with mobile phones, dual screen interaction, through the scan code, can either go directly to the offline store to buy the clothing tried online, or Continue to get the fitting experience on the phone.
“On-line virtual fitting can help brand owners reduce returns and improve user experience; online and offline linkages can activate brand activity and membership.†In the future, good clothing also hopes to access the apparel industry chain. At a deeper level, customization, pre-sales, and even production can all become more scientific with the help of virtual fitting. "Why is the clothes only 4 yards? Why can't we make 20 yards, 50 yards?"
燑br>
Standard Recovery Stud Diode are mainly used for rectifying and switching
It has positive pressure reduction (0.4v -- 1.0v), short reverse recovery time (2-10ns nanosecond), large reverse leakage current, low pressure resistance, generally lower than 150V, and is used in low voltage situations.
Diodes are electronic devices that have two electrodes that allow only a single current to flow in a single direction.Varicap bond (bond) is an electronic adjustable capacitor.Most diodes have a current orientation that we call reading "rectifier".The most common function of a diode is to allow only the current to pass in one direction (called forward bias) and to block in the opposite direction (called reverse bias).
Standard Recovery Stud Diode,Standard Recovery Diode,400V Standard Recovery Diode,Stud Diode
YANGZHOU POSITIONING TECH CO., LTD. , https://www.pst-thyristor.com