The Associated Press美联社

原文阅读：

ECCV 2022 | Dr. Tiancheng Zhao on Behalf of the OmLab Team Reports Two Championships in the first ODinW Challenge

Recently, Dr. Zhao Tiancheng, the director of the Om Artificial Intelligence Research Center of the Binjiang Institute of Zhejiang University and chief scientist of Hangzhou Linker Technology, along with the OmLab team, received championships of the Full-Shot, the Few-Shot, and fourth place in the Zero-Shot competition in the ECCV 2022 ODinW Challenge. Dr. Zhao was invited to deliver a spotlight talk to introduce the advancement and innovation of the OmDet—— a new vision-language object detection framework. 

ECCV (European Conference on Computer Vision) is one of the three most reputable global conferences in the field of computer vision. The ODinW (Object Detection in the Wild) Challenge, hosted by Microsoft Research, aims to advance the ability of computer vision AI for open-world and open setting visual recognition.

Dr. Zhao was invited to participate in the Spotlight Sessions and published the report OmDet: Language-Aware Object Detection with Large-Scale Vision-Language Multi-dataset Pre-training, which received a lot of attention due to the superior performance of the proposed OmDet method 

Object Detection (OD) is an fundamental task in computer vision, widely used in intelligent video monitoring, industrial inspection, robotics, etc. AI vision gains its popularity in both academia and industry and meanwhile faces an increasing innovation difficulty. The classic OD research focused on improving detector networks to achieve higher accuracy and lower latency by using fixed output label sets (e.g., 80 categories in MSCOCO). OmDet, a new object detection framework based on VLP (Visual Language Pre-training), was proposed by the OmLab team. The contiual learning method investigates whether the detector can gradually learn from many OD datasets with an increasing visual vocabulary, and ultimately achieve open vocabulary detection abilities. 

Tested in COCO, Pascal VOC, Wider Face, and Wider Pedestrian OD datasets, OmDet was proven not only to be able to learn from all datasets without label collisions but also has stronger performance than a single dataset detector due to its knowledge sharing between tasks. 

Based on the previous results, the OmLab team conducted another research on larger vocabulary pre-training, using an OD dataset with 20 million images and 4 million unique text labels, which includes manual labeling and pesudo labeling. The resulting model is evaluated on the recently proposed ODinW dataset, which covers 35 different OD tasks in various fields. 

The research shows that zero/few-sample learning and parameter efficiency can be improved effectively by expanding vocabulary through multi-dataset pre-training. 

OmDet achieves state-of-the-art performance on a series of downstream tasks.  In the future, OmDet can be improved with effective task sampling strategies and diversified multimodal datasets. Improve its multilingual understanding abilities and prompt tuning performance are also on the research agenda.

原文阅读：

IDC Market Glance: Hangzhou Linker Technology was Selected as a Representative Manufacturer in the Multimodal Large Model of the Chinese AI Market  

Recently, IDC, a well-known international market research institution, released the "Market Glance: An Overview Report of Chinese AI Large Model Market". Hangzhou Linker Technology is selected as a representative manufacturer in the field of the multimodal large model due to the outstanding performance of OmModel——an industry-leading multimodal AI system.

The report shows an overview of China's AI large model market by selecting major technical service providers in different market segments. The main segments include infrastructure providers, AI development platform providers, AI algorithm providers and industry solution providers. This map helps the Chinese enterprises to understand the development of large models and the leading players in each market segment. 

Dr. Tiancheng Zhao, Chief Scientist of Hangzhou Linker Technology, indicates that as Artificial Intelligence empowers the substantial economy into a new phase, enterprises usually face the situation of limited data resources, poor model generalization ability, and scarcity of high-level talents. AI's large model, as one of the best ways to solve the above problems, can greatly reduce the technical threshold and development cost for enterprises to embrace AI. Large models show superior pattern recognition and reasoning abilities which can help AI technology to tap into the potential value of the industry. At present, AI large models are playing an important role in natural language processing, computer vision, robotics, and other fields of applications. 

Hangzhou Linker Technology with its world's leading multimodal large model technology won two championships of ODinW（Object Detection in the Wild) at one of the most reputable AI conference, ECCV. Also, Linker Technology is one of the platform providers of the first national AI innovation application competition hosted by the Ministry of Industry and Information Technology in 2022. It is the leading technology provider in the field of multimodal pre-training large models in the industry. 

OmModel is a multimodal (vision and language) model that is pre-trained on billions of images, text and videos. The system learns to align linguistic information with visual concepts, and achieve zero-shot and few-shot visual recognition abilities and improve prediction accuracy by 30%. 

In order to empower enterprise users with OmModel to solve real-world challenges, Hangzhou Linker Technology released OmVision Platform, a SaaS AI platform that provides a one-stop solution for AI app building and deployment. It provides a brand-new way for developers to create smart apps that features with its no-code and zero-shot abilities. This platform redefined the production process of computer vision applications through the powerful ability of OmModel’s pre-trained large models. 

Pre-training large models have become an inevitable path for AI to further revolutionize the industry. Hangzhou Linker Technology will continuously explore advanced technology innovation and application, taking technology as the core and application as the guide to provide AI momentum for the digital intelligent upgrade in various industries with its core technology ability. 

DIGITAL JOURNAL

Hangzhou Linker Technology Passed the Evaluation of the Ministry of Industry and Information Technology 

原文阅读：

Hangzhou Linker Technology Passed the Evaluation of the Ministry of Industry and Information Technology 

Recently, the OmModel of Hangzhou Linker Technology passed the large-scale pre-training model system evaluation of the Ministry of Industry and Information Technology of China and obtained two S-level and three A+ reviews in 5 capability evaluation metrics, nine top-level and two advanced evaluations in 11 platform maturity evaluations. As a result, OmModel has become the first large model in China that passes this evaluation. 

Dr. Tiancheng Zhao, Chief Scientist of Hangzhou Linker Technology, indicated that OmModel, as a new infrastructure in the AI field, solved the problems of data resources shortage and long R&D cycles in the AI implementation process. Compared with the traditional development mode, large models require less labeled data and shorter development cycles, which can effectively reduce the costs for the development and application of AI industrialization with the support of low-code development platforms. 

The China Academy of Information and Communications Technology (CAICT) has established AIAE, an authoritative evaluation system for AI innovative application technology and product services, and formulated the "Large-scale Pre-training Model System Capability Assessment Method". OmModel of Hangzhou Linker Technology was evaluated and passed the assessment, which includes two parts: technical service capability and system function, and comprehensively evaluating the function. 

Through the multimodal AI algorithm of large-scale self-supervised learning, OmModel integrates the language and visual modal understanding of "text + image + video" to complete large-scale pre-training on hundreds of millions of pictures, tens of thousands of videos, and billions of images and texts. The model learns to fuse multi-modal information, and obtains more accurate visual AI models with a smaller number of labeled samples, successfully coping with tens of millions of scenarios with a single system.  

Excellent Technical Capabilities 

As a Language Augmented Visual Model, OmModel supports few-shot learning and multimodal fusion, which can use multimodal co-learning algorithms to quickly perform AI modeling based on a small amount of annotated data. It can perform multimodal learning through alignment and conversion across different modalities and perform open-set object detection, outlier detection, image classification and etc. s. Image retrieval, visual question answering, visual common sense reasoning, semantic reference detection, and zero-sample multimodal can also be supported. 

High Platform Maturity 

Hangzhou Linker Technology launched the "Visual AI Application Development Factory" OmVision Studio based on OmModel, which reshapes visual algorithms production with a new process of "0-sample cold start, small-sample training, and online tuning". With data access, data annotation, model training, model fine-tuning, model compression, and other functions, it fulfills clients’ R&D needs at different stages and situations, providing a simple development process, and visualized implementation. At present, OmModel large model and system platform services have been applied in many fields such as radio and television media, energy and power, smart supervision, and government governance, promoting digital transformation in many enterprises. 

Hangzhou Linker Technology will continuously promote the improvement of the pre-trained large models' implementation in technology development, product application, deployment services, etc., to adapt the industrial markets for AI practicality, adaptability, and inclusiveness.