Pristupačnost
The ACDC Challenge 2022 is a major competition in road-scene understanding. The challenge has aimed at evaluating semantic segmentation performance in adverse visual conditions, such as fog, nighttime, rain, and snow. The challenge was organized by researchers from the ETH University of Zurich.
Over 80 submissions from all over the world were received before the challenge deadline in June 2022. The submission of our colleague Matej Grcić, Msc, has attained the best performance in two out of four Tracks of the challenge: T2 - Supervised semantic segmentation in adverse conditions, and T4 - Uncertainty-aware semantic segmentation. His submission also won in 2/4 subtracks of T3 - Supervised semantic segmentation in individual adverse conditions. Matej has delivered an invited presentation of his work at the workshop Vision for all Seasons: Adverse Weather and Lighting Conditions, which was held in conjunction with the CVPR 2022 conference in New Orleans.
The remaining text brings some more information about Matej, his method and a few photos from the competition.
Matej Grcić is employed on the project KK.01.2.1.02.0119 A-Unit co-funded by the European Regional Development Fund and Gideon Brothers. Matej collaborates on the ADEPT project funded by the Croatian Science Foundation and participates in courses Computer Architecture 2, Design Forms in Programming and Deep Learning 1. His method is based on the doctoral dissertations of Ivan Krešo and Marin Oršić. Matej was assisted by Ivan Grubišić and other members of the D307 team. This work was carried out in collaboration with the College for Information Technologies (VSITE), which provided access to a hardware platform with 8 NVIDIA Tesla V100 32GB GPUs.
