Home / Press Releases / Severstal launches Kaggle competition for the use of computer vision in production

Severstal launches Kaggle competition for the use of computer vision in production

Press Releases
463
0

Severstal Digital LLC (part of PJSC Severstal) has completed the development of the first version of the neural network for detecting surface defects on sheet metal in the Metal Processing Shop No. 2 (TsOM-2) of the flat-rolled products production of the Cherepovets Steel Works (CherMK).

The company has already launched a pilot project to confirm the accuracy of the model as one of the elements of the system for automatic certification of the quality of the final product. Analysis of the first results of the model's operation shows a significantly higher accuracy of defect detection compared to the previously used industrial hardware complex for visual inspection.

In order to compare the parameters of the model developed by specialists from the Directorate for Technical Development and Quality (DTRK) and Severstal Digital LLC , solutions that the world's best Data Sciece teams can offer, as well as to further improve this model, the company is launching a competition Kaggle <-! dle_leech_end -> - the world's largest platform for pomashinnomu training events (included in the Google group of companies). The competition started on July 26, 2019 and will last 3 months.

“Kaggle has already established itself as the best platform for finding the most perfect solutions in the field of digital production. At different times, competitions on the Kaggle platform were held by a number of leading foreign and Russian high-tech companies. About 9 thousand teams took part in one of the last competitions. Taking into account the specifics of Severstal's task, we expect that about 1-3 thousand Data Sciece specialists from all over the world will take part in our competition. The quality of the solutions proposed by the participants will be assessed using the Mea Dice Coefficiet metric. Thus, the decisions of the participants will be assessed not only by the accuracy of the classification of the detected metal defects, but also by how correctly these defects were localized, their shape and surface area were determined, "commented Igor Bardintsev, director of digital technologies development at Severstal.

The prize fund is a total of $ 120 thousand: for the first place the winner will receive $ 40 thousand, for the second - $ 25 thousand, the third - $ 15 thousand for the fourth and fifth places - $ 10 thousand each. In addition, there is a separate prize of $ 20 thousand for the fastest solution .

“Allocation of an additional prize for the fastest solution will allow us not only to encourage the creation of the most accurate models, but also to obtain solutions that are optimized for speed of execution while maintaining acceptable quality,” says Igor Bardintsev.

Сomments
Добавить комментарий
Комментарии (0)
To comment
Войти с ВК Войти с ФБ Войти с Яндекс
Sign in with:
Войти с ВК Войти с ФБ Войти с Яндекс