Design Patent Image Classification Challenge
1. The purpose and significance of the competition
Design patent search can help users find the design patent they need based on requirements. Especially when companies are developing new products, they can avoid duplication of development through design patent search. At the same time, they must also use patent search to circumvent essential patents. An important part of patent search includes the classification of images. Therefore, strengthen the classification of design patent images is the research of intelligent retrieval system, which has strong practical value in the field of intellectual property rights. At present, the research on image classification and retrieval is developing rapidly, and there are many domestic and foreign researchers, and the article methods are constantly emerging. However, most of the research is mainly focused on the algorithm level, trying to solve the problems of low accuracy, recall and slow query speed faced in image classification and retrieval models. However, there is currently a lack of publicly available data on design patents, because patent data are generally held by official agencies and cannot be easily obtained by people.
In view of the above situation, this competition will provide a benchmark for the performance evaluation of the classification of clothing design patent images for the general public. The benchmark can be used to test and evaluate classification-related algorithms as a benchmark for public standards. At the same time, it can also promote research in the field of intellectual property and promote the application of artificial intelligence and big data technology in the field of intellectual property.
Organizer: Guangdong Provincial Key Laboratory of Intellectual Property and Big Data
Guidance: The 4th Chinese Conference on Pattern Recognition and Computer Vision (PRCV 2021)
3. Task introduction
The main task of this competition is to classify the design patent images. The participating teams design related algorithms according to the characteristics of the design patent images, use the clothing design patent image data set given by the organizer for model training, and finally the F1-score is used as an evaluation index to measure the performance of the participating team's algorithm model.
4. Competition details and registration entry
- June 01-June 30：Release Competition question: The organizing committee announces the competition task, entry method and evaluation method on the website
- June 01-June 30：Registration: Team registration and qualification confirmation
- July 01-June 10：Release data set: The organizing committee completes the preparation of training samples and verification samples on the website and announces to the participating teams. Each team downloads via the Internet, and conducts algorithm debugging in the team’s own software and hardware environment
- July 11-September 10：Start competition: Each team completes the task within the specified time and uploads the result to the specified location in the specified format
- September 11-September 25：Results evaluation: The organizing committee announces the test data set of the competition, and the judges verify the results of each team's program execution according to the prescribed evaluation method, evaluate the results, and compare the ranking of each team
- October 29-November 01：During the 4th Chinese Conference on Pattern Recognition and Computer Vision (PRCV 2021), a workshop will be held to introduce competition methods and award prizes
6. Contact information
Competition exchange group of QQ: 828450521