On December 20th, Leyard Virtual Dynamic officially released the "Lydia" action model on the AI results release day. Partners such as HUAWEI CLOUD, JD Technology, iFLYTEK, e Cloud, Intel, and Amazon Cloud attended the press conference to witness the historical moment of the AI era.
Liu Yaodong, CMO of Leyard Group and CEO of Virtual Dynamic Node, released the Lydia action model on the spot, which is a professional capability model independently developed by Leyard Virtual Motion Node, which has core capabilities such as "cognition, perception, and generation" of spatial action data, and takes the lead in exploring a new model of AIGC in the field of spatial computing and action generation on a global scale.
The bottleneck in the development of large models in the AI era is no longer the development of algorithms, but the accumulation of data. Leyard Virtual Dynamic Point has accumulated a large amount of motion data in the field of spatial computing and motion capture for many years, which is incomparable to other AI large model companies, and it is also the core reason why Leyard Virtual Dynamic Point can develop large motion models.
The parameters of the Lydia action model include two parts: the basic data parameters and the core action data parameters. In terms of basic data parameters, Lydia refers to today's mainstream large model datasets, and has reached the scale of 10 billion sets in the process of training the model for natural language understandingIn terms of core action data parameters, it has many years of data accumulation of OptiTrack optical motion capture, data precipitation of unmarked point motion capture, and action data refinement in the Internet public**. With the accumulation of three kinds of action data, Leyard Virtual Moving Point successfully trained the action data of the Lydia large model.
The so-called "fierce competition" of today's large-scale model track is inevitably faced with the problem of not landing and product differentiation not being high. The virtual moving point to make a large model is not based on the current trend, but confidence is the "unintentional willow" that has been deeply cultivated in the professional field for many years.
In the field of algorithms, we can't roll up the big factories - no matter how powerful the individual algorithm engineers are, there may be changes, even the core team of Open AI has changed," Liu Yaodong said in a conversation with Leifeng.com, "The algorithm is like this, and the computing power is also easy to migrate, but it is the data that is not easy to migrate." That's what we've been doing at Optitrack for so many years. With this data, we also found an opportunity to make a big model. It can be said that the massive data accumulation of Optitrack has made the birth of Lydia a a matter of course, and it has also become our confidence and confidence in exploring the field of large models."
Liu Yaodong said in his sharing: "Lydia is currently one of the few large models in the world that truly realizes action generation. Different from other general models, the Lydia action model is a professional capability model, which has different capabilities and application fields, complementing each other's capabilities and scenarios. Lydia enables cognitive understanding of action data in the field of spatial computing, enabling efficient action generation. At the same time, Lydia can be connected and called by other software platforms, so that the action files generated by it can be commercialized by ecosystem partners. ”
At the launch site, Cui Chao, technical director of Leyard Virtual Dynamic Point, first demonstrated the cognitive understanding and description ability of the Lydia large model for action data files, which can greatly improve the efficiency of the AI data annotation industry and provide a guarantee for the continuous training of large models on massive dataSecondly, a simple action, a set of complex actions and a series of characteristic actions were continuously generated by the Lydia large modelFinally, the data connection and invocation capabilities of the Lydia large model and the mainstream rendering engine Unity platform were demonstrated, realizing the commercial closed loop of the action large model and other tool platforms.
Lydia and OptiTrack are two technology products with different capabilities. The Lydia model focuses on solving industry scenarios where the efficiency of motion data acquisition is more important, while OptiTrack optical motion capture is better at industry scenarios where accuracy requirements are more important. In short, Lydia focuses on efficiency and OptiTrack focuses on precision, and the two work together to operate efficiently and create value for customers with different needs.
For example, in industries such as film and television, e-commerce, animation production, and games, Lydia can greatly improve the production efficiency of the industry and create greater value for the industry with its AIGC generation capabilities.
So, with Lydia's strong entry into the game, will the business of Virtual Movepoint be affected by the AIGC of the Tiger Downhill?Is it "left hand to right hand" to make a large model with virtual moving points?
Liu Yaodong thought very clearly about this problem, and he said to Leifeng.com: "On Optitrack, its project will not be replaced by Lydia, because the main direction of Optitrack in the future is in areas that require extremely high accuracy - such as medical treatment, we use optical motion capture positioning to operate medical surgical manipulators, and the accuracy error has been less than + -0."1mm, an area that Lydia can't solve;In some fields that do not require high accuracy, but require high efficiency, the advantage of Lydia is that it can greatly improve the cost performance. ”
From this point of view," said Liu Yaodong, "Lydia and Optitrack are hardly the same track, and they can both meet the needs of their respective applicable scenarios."
Bow down and forge ahead. In the context of the AI era, AIGC has become the core force of industry change. In the future, Leyard Virtual Dynamic will continue to improve Lydia's technical capabilities, and work with many partners to build a "professional capability action model ecosystem" to empower thousands of industries.