From identifying oracle bones to identifying the more difficult Zhong Dingwen , the artistic accomplishment of writing poetry and painting has also been greatly improved; in the face of online fraud traps, the "anti-fraud blind box" layer by layer "unblocking" is smarter... 2022 The World Artificial Intelligence Conference kicked off today. Let us take a look. After a year of absence, what new capabilities has artificial intelligence (AI) gained?
Highlight 1: AI recognition of Western Zhou Dynasty Zhong Dingwen
Do you still remember the popular Oracle recognition at the World Artificial Intelligence Conference last year? At this year's conference, intelligent text recognition technology was applied to the more difficult recognition of Zhongdingwen (bronze inscriptions).
In the Hehe Information booth, "Seat C" is an ancient tripod. The uneven inner wall of the ancient tripod is engraved with two "incomprehensible" words. The reporter pressed the shooting button, and the camera in Gudingli focused on the text, took a picture, and projected the picture onto the big screen. Then, something magical happened. Without human intervention, the characters inside the tripod were flattened from the concave state and translated into simplified Chinese characters. Since then, ancient Chinese sentences that were originally connected together, such as "Ke Yue Mu Mu Zhenwen and Master Huafu's 悂歲氒心 tranquility in Youshu Zheye De", were also separated after automatic processing by the "AI sentence segmentation" function. According to the staff, the translation of this tripod article roughly means the origin story of the grandfather in the family who got the tripod. In Zhong Dingwen's processing of
, the key "skill points" of intelligent text recognition are included, including intelligent image processing represented by "bend correction", complex scene text recognition based on deep learning, and natural language processing (NLP). wait. According to the staff, these technologies have many application scenarios in real life. “At this stage, image processing and have problems such as serious degradation of document image quality, difficulty in text detection and layout analysis, and low non-qualified text recognition rate in different scenarios. , intelligent image processing technology can efficiently and accurately process document images in complex scenes."
Highlight 2: Magic Pen AI can turn a word into a picture
AI can write poems, paintings, and text? In the AI creative experience area specially designed by Baidu Feipiao, the reporter experienced "Everyone is an artist". He entered "Walden Pond" on his mobile phone and clicked "Generate Now", and the painting process began to be displayed on the big screen. Generate a beautiful painting just like printing and downloading. Click Apply again, and the living room frame background, pillows, T-shirts and other scenes will appear on the screen.
can not only create brand new paintings, this "magic pen Ma Liang" can also "complete" Huang Gongwang's handed down masterpiece " Residence in the Fuchun Mountains ". Because the middle part of this famous painting has been damaged, the two existing fragments cannot give a full view of it. The reporter experienced the "virtual completion" of "Dwelling in the Fuchun Mountains" on site. He randomly drew a few strokes of landscapes and houses between the two fragments, and clicked "AI Generate" to connect the landscapes in one second, and based on For the research and data accumulation of famous paintings, the AI-completed parts can even be consistent with the style of the existing paintings, and the landscape context in the paintings is also harmonious and smooth. Why can
"one word make a picture" and "complete a famous picture" be easily realized? The staff revealed that this comes from the AI art and creative assistance platform "Wenxin·Yige" just released by Baidu , and the latter has realized product innovation based on the Wensheng diagram system of flying paddle Wenxin large model. The learning and creation capabilities of artificial intelligence in the field of art are vigorously refreshing our understanding, and at the same time, it also provides greater imagination for the integration and innovation of technology, art and culture.
Highlight 3: "Anti-fraud blind box" "Unblocking" layer by layer
Walking into the Ant Group booth, an "anti-fraud blind box" attracted the attention of many people. The reporter randomly selected a blind box, and what popped up was a vivid campus loan video. At the end of the video, a QR code popped up prompting a transfer. There is only one step left before the scammer succeeds. At this time, technological means can help. When the reporter simulated scanning the QR code and inputting the password to transfer money, his mobile phone immediately received a customer service call from the anti-fraud center. The "AI wake-up robot" on the other end of the phone would carefully inquire about the transaction product information, the name of the trading platform, etc. , and reminded reporters of the risk of being deceived.
If consumers insist on choosing to pay, technicians have also newly developed the "15-minute cooling-off period" and "24-hour delayed payment" functions. The former actively suspends risky transactions to give users a time window to verify and choose again; the latter The latter is equivalent to the transaction regret period, which means that users still have the possibility to recover risk funds within 24 hours. The principle of the
"anti-fraud blind box" interactive product is derived from the "intelligent risk perception and response joint anti-fraud system" independently developed by Ant Group. According to the staff, this intelligent anti-fraud system is based on technological breakthroughs such as interactive risk control, full-picture risk control, and trusted artificial intelligence. It can realize risk control automatic driving and automatic anomaly sensing, identification and intervention during the incident, and intelligent disposal after the incident. Cross-industry joint prevention and control has provided security protection for more than 1 billion users around the world and hundreds of millions of transactions every day.
Highlight 4: The "first step" in socializing in the metaverse
Can you socialize without showing your face? This is possible with an app called Soul. The reporter saw at the Shanghai Anydoor Technology Co., Ltd. booth in the WAIC Yuanverse exhibition area that users wearing various masks and hoods were sharing their photos and short videos. Some were cartoon girls, and some were Peking Opera masks. , as well as animal images such as unicorns and tiger cubs. What’s interesting is that these “masks” can vividly restore the expressions and demeanor of real people.
“These are virtual images made by users themselves,” the staff told reporters. The “face-pinching” function is based on the “NAWA” engine independently developed by Lingxi App. The ability to vividly present the user's expression is because it uses a large number of algorithms when collecting information and can recognize the user's rich expressions. "For example, NAWA can finely identify micro-expressions such as blinking, sticking out the tongue, and bulging the cheeks." and link up."
Highlight 5: The city has a "twin brother"
Digital twin is the virtual and real mapping and drive control between the physical world and the digital world. At the booth of Qianxun, it was demonstrated for the first time that it has precise time and space capabilities. Li's digital twin products include the 3D map engine "Qianxun Shujing" for digital twins, the road intelligent inspection system "Qianxun Chiguan", the digital twin infrastructure management platform "Qianxun Twin World", etc., which can Help build a "digital twin" city that accurately maps the real environment in real time.
reporters saw at the scene that the "Qianxun Twin World" on display this time restores the location, posture and other spatio-temporal information of the infrastructure based on a high-precision unified spatio-temporal benchmark, and supports the simultaneous projection of real-time sensing data in the virtual world. Managers can trace back the people, vehicles, objects and events that can be sensed in any time and space scene to complete the refined management of the entire life cycle of the infrastructure.
"More refined digital twin capabilities are the basis for realizing spatiotemporal intelligence." Chen Jinpei, CEO of Qianxun Location, said that only with accurate mapping can physical cities and digital cities be closely integrated and interact in two directions, thereby realizing various spatiotemporal intelligent applications.
also demonstrated smart transportation, smart driving, digital city and other cases on site. For example, in the field of road maintenance, the traditional inspection process usually has problems such as low automation, low accuracy of algorithm recognition, duplication of data, and limited collection coverage. Through "Qianxun Chiguan", only one driver can collect three-lane road conditions, support edge-side intelligent identification of typical road diseases and equipment and facilities, and the identification accuracy is better than 90%, which can greatly improve daily inspections. Improve operating efficiency, reduce operating labor costs, and provide auxiliary decision-making support for road inspections.
Author: Xu Jinghui Zhang Tianchi
Editor: Zhu Wei
