There comes the ICCV 2021, which has been waiting for us. With the corporate mission of “Helping people eat better, live better”, meituan will focus on the application of computer vision technology in the field of large-scale fine-grained food analysis.

From 19:00 to 22:30, October 16, Beijing time, Meituan Visual Intelligence Department will join hands with the Institute of Computing Science of Chinese Academy of Sciences, Beijing Zhiyuan and The University of Barcelona to exchange and discuss the theme of “LargeFineFoodAI” with well-known experts and scholars in the field of food computing from all over the world.

The agenda of the seminar will be divided into the following three parts. Due to the impact of the epidemic, the whole process will be conducted online (please refer to the participation method at the bottom of the article).

01 Invited Talk

As one of the three top conferences of computer vision, ICCV international Computer Vision Conference takes “LargeFineFoodAI” as the theme and focuses on fine-grained recognition and retrieval of large-scale food images for the first time. In the highly anticipated sharing session, we have invited three leading experts in the industry to bring the latest theoretical and practical results on computer vision applications in the food field.

Food is an important source of physical health and mental well-being for everyone, but an individual’s dietary preferences may not match his physical fitness. For most people, striking a balance between healthy eating and happy eating is very difficult. Ideally, a food model should be built for each person to customize their diet. Therefore, this sharing will focus on how to build a personal food model and food atlas.

A new food logging tool, FoodLog Athl, is available for diet-related healthcare and dietary assessment services. From the perspective of dietitians or monitoring users, the tool can support food image recognition, nutritional diet evaluation, food nutritional value measurement and other functions. In addition, key scientific findings such as the role of the tool before and after COVID-19 and changes in relevant food statistics will be presented.

Neural network has become one of the most powerful predictive systems. The Bayesian paradigm of deep learning takes probability as the learning objective of neural network architecture and parameters, and quantifies the predicted uncertainty through posterior distribution. In this presentation, we will explore why uncertainty estimation is needed, how it can be modeled and measured, and to further demonstrate its application value, we will explore how food identification can be modeled using uncertainty.

02 Challenge Report

The “Large-scale food image recognition and Retrieval” challenge organized by the symposium also attracted many strong teams from home and abroad, including Tsinghua University, University of Science and Technology of China, Nanjing University of Science and Technology, University of Barcelona, Nanyang Technological University of Singapore; A total of 143 teams from home and abroad, including Alibaba, Shenlan Technology, OPPO and Jubilee, participated in the competition.

The competition is divided into two categories: large scale food image fine granularity recognition and large scale food image fine granularity retrieval, which are selected according to the final results and submitted technical schemes. The following winning teams will also present their competition results and technical solutions.

Table 1. The ranking list of the top-3 teams in the “Large-scale fine-grained food recognition” challenge at ICCV 2021

Table 2. The ranking list of the top-3 teams in the “Large-scale fine-grained food retrieval” challenge at ICCV 2021

Bonus: Data sets are constantly open

In this competition, we presented a dataset containing over 1,000 fine-grained food categories and over 500,000 images, including Chinese and Western food. With the help of food experts, a unified food ontology was constructed by combining and adapting the existing food classification system. The number of images in each category is within the range of [153; 1999], which shows a greater category imbalance compared with the existing food data set, and also brings greater challenges for recognition and retrieval! Through this symposium and competition, we will continue to make this data set available to advance the application of computer vision in the field of food analysis.

Prize interactive

In the results report session of the challenge on October 16, online audience participating in the interactive questioning will have the opportunity to win the following prizes. Looking forward to your questions and challenges!

03 Oral Presentation

According to ICCV official news, a total of 6,236 papers were submitted this year, an increase of about 50% compared with last year; In the end, 1617 papers were accepted, with an acceptance rate of 25.9%. The LargeFineFoodAI workshop was solicited on the theme of food analysis and received high-quality paper feedback, culminated in the acceptance of two papers from Carnegie Mellon University and Purdue University. The authors of the two papers will also be invited to give excellent reports and interpretations.

When we use computer vision technology to re-explore food, when the frontier and traditional collision, how to let everyone eat better, more scientific, more healthy? Look forward to working with us to find out!

Please long press or scan the qr code above, reply “LargeFineFoodAI”, you will automatically join LargeFineFoodAI2021 technical exchange group.

Full agenda: foodai-workshop.meituan.com/foodai2021….